Spaces:
Runtime error
Runtime error
update token map
Browse files- app.py +202 -100
- models/attention.py +20 -8
- models/region_diffusion.py +222 -31
- models/unet_2d_blocks.py +244 -59
- utils/attention_utils.py +147 -25
- utils/richtext_utils.py +8 -8
app.py
CHANGED
@@ -22,18 +22,17 @@ from share_btn import community_icon_html, loading_icon_html, share_js, css
|
|
22 |
help_text = """
|
23 |
If you are encountering an error or not achieving your desired outcome, here are some potential reasons and recommendations to consider:
|
24 |
1. If you format only a portion of a word rather than the complete word, an error may occur.
|
25 |
-
2.
|
26 |
-
3.
|
27 |
-
4. Consider using a different seed.
|
28 |
"""
|
29 |
|
30 |
|
31 |
canvas_html = """<iframe id='rich-text-root' style='width:100%' height='360px' src='file=rich-text-to-json-iframe.html' frameborder='0' scrolling='no'></iframe>"""
|
32 |
get_js_data = """
|
33 |
-
async (text_input, negative_prompt, height, width, seed, steps, guidance_weight, color_guidance_weight, rich_text_input) => {
|
34 |
const richEl = document.getElementById("rich-text-root");
|
35 |
const data = richEl? richEl.contentDocument.body._data : {};
|
36 |
-
return [text_input, negative_prompt, height, width, seed, steps, guidance_weight, color_guidance_weight, JSON.stringify(data)];
|
37 |
}
|
38 |
"""
|
39 |
set_js_data = """
|
@@ -71,9 +70,13 @@ def main():
|
|
71 |
width: int,
|
72 |
seed: int,
|
73 |
steps: int,
|
|
|
|
|
|
|
74 |
guidance_weight: float,
|
75 |
color_guidance_weight: float,
|
76 |
-
rich_text_input: str
|
|
|
77 |
):
|
78 |
run_dir = 'results/'
|
79 |
# Load region diffusion model.
|
@@ -88,7 +91,7 @@ def main():
|
|
88 |
# parse json to span attributes
|
89 |
base_text_prompt, style_text_prompts, footnote_text_prompts, footnote_target_tokens,\
|
90 |
color_text_prompts, color_names, color_rgbs, size_text_prompts_and_sizes, use_grad_guidance = parse_json(
|
91 |
-
json.loads(text_input)
|
92 |
|
93 |
# create control input for region diffusion
|
94 |
region_text_prompts, region_target_token_ids, base_tokens = get_region_diffusion_input(
|
@@ -108,7 +111,7 @@ def main():
|
|
108 |
# get token maps from plain text to image generation.
|
109 |
begin_time = time.time()
|
110 |
if model.attention_maps is None:
|
111 |
-
model.
|
112 |
else:
|
113 |
model.reset_attention_maps()
|
114 |
plain_img = model.produce_attn_maps([base_text_prompt], [negative_text],
|
@@ -116,27 +119,38 @@ def main():
|
|
116 |
guidance_scale=guidance_weight)
|
117 |
print('time lapses to get attention maps: %.4f' %
|
118 |
(time.time()-begin_time))
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
123 |
color_obj_masks = [transforms.functional.resize(color_obj_mask, (height, width),
|
124 |
interpolation=transforms.InterpolationMode.BICUBIC,
|
125 |
antialias=True)
|
126 |
for color_obj_mask in color_obj_masks]
|
127 |
text_format_dict['color_obj_atten'] = color_obj_masks
|
128 |
-
model.
|
129 |
|
130 |
# generate image from rich text
|
131 |
begin_time = time.time()
|
132 |
seed_everything(seed)
|
|
|
|
|
|
|
|
|
133 |
rich_img = model.prompt_to_img(region_text_prompts, [negative_text],
|
134 |
height=height, width=width, num_inference_steps=steps,
|
135 |
-
guidance_scale=guidance_weight,
|
136 |
-
text_format_dict=text_format_dict
|
|
|
137 |
print('time lapses to generate image from rich text: %.4f' %
|
138 |
(time.time()-begin_time))
|
139 |
-
return [plain_img[0], rich_img[0], token_maps]
|
140 |
|
141 |
with gr.Blocks(css=css) as demo:
|
142 |
url_params = gr.JSON({}, visible=False, label="URL Params")
|
@@ -162,6 +176,29 @@ def main():
|
|
162 |
placeholder='Example: poor quality, blurry, dark, low resolution, low quality, worst quality',
|
163 |
elem_id="negative_prompt"
|
164 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
165 |
seed = gr.Slider(label='Seed',
|
166 |
minimum=0,
|
167 |
maximum=100000,
|
@@ -169,15 +206,14 @@ def main():
|
|
169 |
value=6,
|
170 |
elem_id="seed"
|
171 |
)
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
value=0.5)
|
177 |
with gr.Accordion('Other Parameters', open=False):
|
178 |
steps = gr.Slider(label='Number of Steps',
|
179 |
minimum=0,
|
180 |
-
maximum=
|
181 |
step=1,
|
182 |
value=41)
|
183 |
guidance_weight = gr.Slider(label='CFG weight',
|
@@ -206,6 +242,8 @@ def main():
|
|
206 |
with gr.Row():
|
207 |
plaintext_result = gr.Image(
|
208 |
label='Plain-text', elem_id="plain-text-image")
|
|
|
|
|
209 |
token_map = gr.Image(label='Token Maps')
|
210 |
with gr.Row(visible=False) as share_row:
|
211 |
with gr.Group(elem_id="share-btn-container"):
|
@@ -218,181 +256,238 @@ def main():
|
|
218 |
gr.Markdown(help_text)
|
219 |
|
220 |
with gr.Row():
|
221 |
-
|
222 |
[
|
223 |
-
'{"ops":[{"insert":"a "},{"attributes":{"
|
224 |
'',
|
225 |
-
|
226 |
-
|
|
|
227 |
6,
|
228 |
1,
|
229 |
-
None
|
|
|
230 |
],
|
231 |
[
|
232 |
-
'{"ops":[{"insert":"
|
233 |
'',
|
234 |
-
|
235 |
-
|
236 |
-
|
|
|
237 |
1,
|
238 |
-
None
|
|
|
239 |
],
|
240 |
[
|
241 |
-
'{"ops":[{"attributes":{"link":"
|
242 |
-
'
|
243 |
-
|
244 |
-
|
245 |
-
|
|
|
246 |
1,
|
247 |
-
None
|
|
|
248 |
],
|
249 |
]
|
250 |
-
|
251 |
-
|
|
|
252 |
inputs=[
|
253 |
text_input,
|
254 |
negative_prompt,
|
255 |
-
|
256 |
-
|
|
|
257 |
seed,
|
258 |
color_guidance_weight,
|
259 |
rich_text_input,
|
|
|
260 |
],
|
261 |
outputs=[
|
262 |
plaintext_result,
|
263 |
richtext_result,
|
|
|
264 |
token_map,
|
265 |
],
|
266 |
fn=generate,
|
267 |
# cache_examples=True,
|
268 |
examples_per_page=20)
|
269 |
with gr.Row():
|
270 |
-
|
271 |
[
|
272 |
-
'{"ops":[{"insert":"
|
273 |
-
'',
|
274 |
-
|
275 |
-
|
|
|
276 |
6,
|
277 |
-
|
278 |
-
None
|
|
|
279 |
],
|
280 |
[
|
281 |
-
'{"ops":[{"insert":"
|
282 |
-
'',
|
283 |
-
|
284 |
-
|
|
|
285 |
6,
|
286 |
-
1,
|
287 |
-
None
|
|
|
288 |
],
|
289 |
[
|
290 |
-
'{"ops":[{"insert":"
|
291 |
'',
|
292 |
-
|
293 |
-
|
|
|
294 |
6,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
295 |
1,
|
296 |
-
None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
297 |
],
|
298 |
]
|
299 |
-
|
300 |
-
|
301 |
-
label='Footnote examples',
|
302 |
inputs=[
|
303 |
text_input,
|
304 |
negative_prompt,
|
305 |
-
|
306 |
-
|
|
|
307 |
seed,
|
308 |
color_guidance_weight,
|
309 |
rich_text_input,
|
|
|
310 |
],
|
311 |
outputs=[
|
312 |
plaintext_result,
|
313 |
richtext_result,
|
|
|
314 |
token_map,
|
315 |
],
|
316 |
fn=generate,
|
317 |
# cache_examples=True,
|
318 |
examples_per_page=20)
|
|
|
319 |
with gr.Row():
|
320 |
-
|
321 |
[
|
322 |
-
'{"ops":[{"insert":"a
|
323 |
'',
|
324 |
-
|
325 |
-
|
326 |
-
|
327 |
-
|
328 |
-
|
|
|
|
|
329 |
],
|
330 |
[
|
331 |
-
'{"ops":[{"
|
332 |
-
'',
|
333 |
-
|
334 |
-
|
|
|
335 |
9,
|
336 |
-
|
337 |
-
None
|
|
|
338 |
],
|
339 |
[
|
340 |
-
'{"ops":[{"insert":"
|
341 |
'',
|
342 |
-
|
343 |
-
|
344 |
-
|
345 |
-
|
346 |
-
|
|
|
|
|
347 |
],
|
348 |
]
|
349 |
-
gr.Examples(examples=
|
350 |
-
label='Font
|
351 |
inputs=[
|
352 |
text_input,
|
353 |
negative_prompt,
|
354 |
-
|
355 |
-
|
|
|
356 |
seed,
|
357 |
color_guidance_weight,
|
358 |
rich_text_input,
|
|
|
359 |
],
|
360 |
outputs=[
|
361 |
plaintext_result,
|
362 |
richtext_result,
|
|
|
363 |
token_map,
|
364 |
],
|
365 |
fn=generate,
|
366 |
# cache_examples=True,
|
367 |
examples_per_page=20)
|
|
|
368 |
with gr.Row():
|
369 |
size_examples = [
|
370 |
[
|
371 |
'{"ops": [{"insert": "A pizza with "}, {"attributes": {"size": "60px"}, "insert": "pineapple"}, {"insert": ", pepperoni, and mushroom on the top, 4k, photorealistic"}]}',
|
372 |
'blurry, art, painting, rendering, drawing, sketch, ugly, duplicate, morbid, mutilated, mutated, deformed, disfigured low quality, worst quality',
|
373 |
-
|
374 |
-
|
|
|
375 |
13,
|
376 |
1,
|
377 |
-
None
|
|
|
378 |
],
|
379 |
[
|
380 |
'{"ops": [{"insert": "A pizza with pineapple, "}, {"attributes": {"size": "20px"}, "insert": "pepperoni"}, {"insert": ", and mushroom on the top, 4k, photorealistic"}]}',
|
381 |
'blurry, art, painting, rendering, drawing, sketch, ugly, duplicate, morbid, mutilated, mutated, deformed, disfigured low quality, worst quality',
|
382 |
-
|
383 |
-
|
|
|
384 |
13,
|
385 |
1,
|
386 |
-
None
|
|
|
387 |
],
|
388 |
[
|
389 |
'{"ops": [{"insert": "A pizza with pineapple, pepperoni, and "}, {"attributes": {"size": "70px"}, "insert": "mushroom"}, {"insert": " on the top, 4k, photorealistic"}]}',
|
390 |
'blurry, art, painting, rendering, drawing, sketch, ugly, duplicate, morbid, mutilated, mutated, deformed, disfigured low quality, worst quality',
|
391 |
-
|
392 |
-
|
|
|
393 |
13,
|
394 |
1,
|
395 |
-
None
|
|
|
396 |
],
|
397 |
]
|
398 |
gr.Examples(examples=size_examples,
|
@@ -400,15 +495,18 @@ def main():
|
|
400 |
inputs=[
|
401 |
text_input,
|
402 |
negative_prompt,
|
403 |
-
|
404 |
-
|
|
|
405 |
seed,
|
406 |
color_guidance_weight,
|
407 |
rich_text_input,
|
|
|
408 |
],
|
409 |
outputs=[
|
410 |
plaintext_result,
|
411 |
richtext_result,
|
|
|
412 |
token_map,
|
413 |
],
|
414 |
fn=generate,
|
@@ -423,11 +521,15 @@ def main():
|
|
423 |
width,
|
424 |
seed,
|
425 |
steps,
|
|
|
|
|
|
|
426 |
guidance_weight,
|
427 |
color_guidance_weight,
|
428 |
-
rich_text_input
|
|
|
429 |
],
|
430 |
-
outputs=[plaintext_result, richtext_result, token_map],
|
431 |
_js=get_js_data
|
432 |
).then(
|
433 |
fn=lambda: gr.update(visible=True), inputs=None, outputs=share_row, queue=False)
|
|
|
22 |
help_text = """
|
23 |
If you are encountering an error or not achieving your desired outcome, here are some potential reasons and recommendations to consider:
|
24 |
1. If you format only a portion of a word rather than the complete word, an error may occur.
|
25 |
+
2. If you use font color and get completely corrupted results, you may consider decrease the color weight lambda.
|
26 |
+
3. Consider using a different seed.
|
|
|
27 |
"""
|
28 |
|
29 |
|
30 |
canvas_html = """<iframe id='rich-text-root' style='width:100%' height='360px' src='file=rich-text-to-json-iframe.html' frameborder='0' scrolling='no'></iframe>"""
|
31 |
get_js_data = """
|
32 |
+
async (text_input, negative_prompt, height, width, seed, steps, num_segments, segment_threshold, inject_interval, guidance_weight, color_guidance_weight, rich_text_input, background_aug) => {
|
33 |
const richEl = document.getElementById("rich-text-root");
|
34 |
const data = richEl? richEl.contentDocument.body._data : {};
|
35 |
+
return [text_input, negative_prompt, height, width, seed, steps, num_segments, segment_threshold, inject_interval, guidance_weight, color_guidance_weight, JSON.stringify(data), background_aug];
|
36 |
}
|
37 |
"""
|
38 |
set_js_data = """
|
|
|
70 |
width: int,
|
71 |
seed: int,
|
72 |
steps: int,
|
73 |
+
num_segments: int,
|
74 |
+
segment_threshold: float,
|
75 |
+
inject_interval: float,
|
76 |
guidance_weight: float,
|
77 |
color_guidance_weight: float,
|
78 |
+
rich_text_input: str,
|
79 |
+
background_aug: bool,
|
80 |
):
|
81 |
run_dir = 'results/'
|
82 |
# Load region diffusion model.
|
|
|
91 |
# parse json to span attributes
|
92 |
base_text_prompt, style_text_prompts, footnote_text_prompts, footnote_target_tokens,\
|
93 |
color_text_prompts, color_names, color_rgbs, size_text_prompts_and_sizes, use_grad_guidance = parse_json(
|
94 |
+
json.loads(text_input))
|
95 |
|
96 |
# create control input for region diffusion
|
97 |
region_text_prompts, region_target_token_ids, base_tokens = get_region_diffusion_input(
|
|
|
111 |
# get token maps from plain text to image generation.
|
112 |
begin_time = time.time()
|
113 |
if model.attention_maps is None:
|
114 |
+
model.register_tokenmap_hooks()
|
115 |
else:
|
116 |
model.reset_attention_maps()
|
117 |
plain_img = model.produce_attn_maps([base_text_prompt], [negative_text],
|
|
|
119 |
guidance_scale=guidance_weight)
|
120 |
print('time lapses to get attention maps: %.4f' %
|
121 |
(time.time()-begin_time))
|
122 |
+
seed_everything(seed)
|
123 |
+
color_obj_masks, segments_vis, token_maps = get_token_maps(model.selfattn_maps, model.crossattn_maps, model.n_maps, run_dir,
|
124 |
+
512//8, 512//8, color_target_token_ids[:-1], seed,
|
125 |
+
base_tokens, segment_threshold=segment_threshold, num_segments=num_segments,
|
126 |
+
return_vis=True)
|
127 |
+
seed_everything(seed)
|
128 |
+
model.masks, segments_vis, token_maps = get_token_maps(model.selfattn_maps, model.crossattn_maps, model.n_maps, run_dir,
|
129 |
+
512//8, 512//8, region_target_token_ids[:-1], seed,
|
130 |
+
base_tokens, segment_threshold=segment_threshold, num_segments=num_segments,
|
131 |
+
return_vis=True)
|
132 |
color_obj_masks = [transforms.functional.resize(color_obj_mask, (height, width),
|
133 |
interpolation=transforms.InterpolationMode.BICUBIC,
|
134 |
antialias=True)
|
135 |
for color_obj_mask in color_obj_masks]
|
136 |
text_format_dict['color_obj_atten'] = color_obj_masks
|
137 |
+
model.remove_tokenmap_hooks()
|
138 |
|
139 |
# generate image from rich text
|
140 |
begin_time = time.time()
|
141 |
seed_everything(seed)
|
142 |
+
if background_aug:
|
143 |
+
bg_aug_end = 500
|
144 |
+
else:
|
145 |
+
bg_aug_end = 1000
|
146 |
rich_img = model.prompt_to_img(region_text_prompts, [negative_text],
|
147 |
height=height, width=width, num_inference_steps=steps,
|
148 |
+
guidance_scale=guidance_weight, use_guidance=use_grad_guidance,
|
149 |
+
text_format_dict=text_format_dict, inject_selfattn=inject_interval,
|
150 |
+
bg_aug_end=bg_aug_end)
|
151 |
print('time lapses to generate image from rich text: %.4f' %
|
152 |
(time.time()-begin_time))
|
153 |
+
return [plain_img[0], rich_img[0], segments_vis, token_maps]
|
154 |
|
155 |
with gr.Blocks(css=css) as demo:
|
156 |
url_params = gr.JSON({}, visible=False, label="URL Params")
|
|
|
176 |
placeholder='Example: poor quality, blurry, dark, low resolution, low quality, worst quality',
|
177 |
elem_id="negative_prompt"
|
178 |
)
|
179 |
+
segment_threshold = gr.Slider(label='Token map threshold',
|
180 |
+
info='(See less area in token maps? Decrease this. See too much area? Increase this.)',
|
181 |
+
minimum=0,
|
182 |
+
maximum=1,
|
183 |
+
step=0.01,
|
184 |
+
value=0.25)
|
185 |
+
inject_interval = gr.Slider(label='Detail preservation',
|
186 |
+
info='(To preserve more structure from plain-text generation, increase this. To see more rich-text attributes, decrease this.)',
|
187 |
+
minimum=0,
|
188 |
+
maximum=1,
|
189 |
+
step=0.01,
|
190 |
+
value=0.)
|
191 |
+
color_guidance_weight = gr.Slider(label='Color weight',
|
192 |
+
info='(To obtain more precise color, increase this, while too large value may cause artifacts.)',
|
193 |
+
minimum=0,
|
194 |
+
maximum=2,
|
195 |
+
step=0.1,
|
196 |
+
value=0.5)
|
197 |
+
num_segments = gr.Slider(label='Number of segments',
|
198 |
+
minimum=2,
|
199 |
+
maximum=20,
|
200 |
+
step=1,
|
201 |
+
value=9)
|
202 |
seed = gr.Slider(label='Seed',
|
203 |
minimum=0,
|
204 |
maximum=100000,
|
|
|
206 |
value=6,
|
207 |
elem_id="seed"
|
208 |
)
|
209 |
+
background_aug = gr.Checkbox(
|
210 |
+
label='Precise region alignment',
|
211 |
+
info='(For strict region alignment, select this option, but beware of potential artifacts when using with style.)',
|
212 |
+
value=True)
|
|
|
213 |
with gr.Accordion('Other Parameters', open=False):
|
214 |
steps = gr.Slider(label='Number of Steps',
|
215 |
minimum=0,
|
216 |
+
maximum=500,
|
217 |
step=1,
|
218 |
value=41)
|
219 |
guidance_weight = gr.Slider(label='CFG weight',
|
|
|
242 |
with gr.Row():
|
243 |
plaintext_result = gr.Image(
|
244 |
label='Plain-text', elem_id="plain-text-image")
|
245 |
+
segments = gr.Image(label='Segmentation')
|
246 |
+
with gr.Row():
|
247 |
token_map = gr.Image(label='Token Maps')
|
248 |
with gr.Row(visible=False) as share_row:
|
249 |
with gr.Group(elem_id="share-btn-container"):
|
|
|
256 |
gr.Markdown(help_text)
|
257 |
|
258 |
with gr.Row():
|
259 |
+
footnote_examples = [
|
260 |
[
|
261 |
+
'{"ops":[{"insert":"A close-up 4k dslr photo of a "},{"attributes":{"link":"A cat wearing sunglasses and a bandana around its neck."},"insert":"cat"},{"insert":" riding a scooter. Palm trees in the background."}]}',
|
262 |
'',
|
263 |
+
5,
|
264 |
+
0.3,
|
265 |
+
0,
|
266 |
6,
|
267 |
1,
|
268 |
+
None,
|
269 |
+
True
|
270 |
],
|
271 |
[
|
272 |
+
'{"ops":[{"insert":"A "},{"attributes":{"link":"kitchen island with a stove with gas burners and a built-in oven "},"insert":"kitchen island"},{"insert":" next to a "},{"attributes":{"link":"an open refrigerator stocked with fresh produce, dairy products, and beverages. "},"insert":"refrigerator"},{"insert":", by James McDonald and Joarc Architects, home, interior, octane render, deviantart, cinematic, key art, hyperrealism, sun light, sunrays, canon eos c 300, ƒ 1.8, 35 mm, 8k, medium - format print"}]}',
|
273 |
'',
|
274 |
+
6,
|
275 |
+
0.5,
|
276 |
+
0,
|
277 |
+
6,
|
278 |
1,
|
279 |
+
None,
|
280 |
+
True
|
281 |
],
|
282 |
[
|
283 |
+
'{"ops":[{"insert":"A "},{"attributes":{"link":"Happy Kung fu panda art, elder, asian art, volumetric lighting, dramatic scene, ultra detailed, realism, chinese"},"insert":"panda"},{"insert":" standing on a cliff by a waterfall, wildlife photography, photograph, high quality, wildlife, f 1.8, soft focus, 8k, national geographic, award - winning photograph by nick nichols"}]}',
|
284 |
+
'',
|
285 |
+
4,
|
286 |
+
0.3,
|
287 |
+
0,
|
288 |
+
4,
|
289 |
1,
|
290 |
+
None,
|
291 |
+
True
|
292 |
],
|
293 |
]
|
294 |
+
|
295 |
+
gr.Examples(examples=footnote_examples,
|
296 |
+
label='Footnote examples',
|
297 |
inputs=[
|
298 |
text_input,
|
299 |
negative_prompt,
|
300 |
+
num_segments,
|
301 |
+
segment_threshold,
|
302 |
+
inject_interval,
|
303 |
seed,
|
304 |
color_guidance_weight,
|
305 |
rich_text_input,
|
306 |
+
background_aug,
|
307 |
],
|
308 |
outputs=[
|
309 |
plaintext_result,
|
310 |
richtext_result,
|
311 |
+
segments,
|
312 |
token_map,
|
313 |
],
|
314 |
fn=generate,
|
315 |
# cache_examples=True,
|
316 |
examples_per_page=20)
|
317 |
with gr.Row():
|
318 |
+
color_examples = [
|
319 |
[
|
320 |
+
'{"ops":[{"insert":"a beautifule girl with big eye, skin, and long "},{"attributes":{"color":"#00ffff"},"insert":"hair"},{"insert":", t-shirt, bursting with vivid color, intricate, elegant, highly detailed, photorealistic, digital painting, artstation, illustration, concept art."}]}',
|
321 |
+
'lowres, had anatomy, bad hands, cropped, worst quality',
|
322 |
+
9,
|
323 |
+
0.25,
|
324 |
+
0.3,
|
325 |
6,
|
326 |
+
0.5,
|
327 |
+
None,
|
328 |
+
True
|
329 |
],
|
330 |
[
|
331 |
+
'{"ops":[{"insert":"a beautifule girl with big eye, skin, and long "},{"attributes":{"color":"#eeeeee"},"insert":"hair"},{"insert":", t-shirt, bursting with vivid color, intricate, elegant, highly detailed, photorealistic, digital painting, artstation, illustration, concept art."}]}',
|
332 |
+
'lowres, had anatomy, bad hands, cropped, worst quality',
|
333 |
+
9,
|
334 |
+
0.25,
|
335 |
+
0.3,
|
336 |
6,
|
337 |
+
0.1,
|
338 |
+
None,
|
339 |
+
True
|
340 |
],
|
341 |
[
|
342 |
+
'{"ops":[{"insert":"a Gothic "},{"attributes":{"color":"#FD6C9E"},"insert":"church"},{"insert":" in a the sunset with a beautiful landscape in the background."}]}',
|
343 |
'',
|
344 |
+
5,
|
345 |
+
0.3,
|
346 |
+
0.3,
|
347 |
6,
|
348 |
+
0.5,
|
349 |
+
None,
|
350 |
+
False
|
351 |
+
],
|
352 |
+
[
|
353 |
+
'{"ops":[{"insert":"A mesmerizing sight that captures the beauty of a "},{"attributes":{"color":"#4775fc"},"insert":"rose"},{"insert":" blooming, close up"}]}',
|
354 |
+
'',
|
355 |
+
3,
|
356 |
+
0.3,
|
357 |
+
0,
|
358 |
+
9,
|
359 |
1,
|
360 |
+
None,
|
361 |
+
False
|
362 |
+
],
|
363 |
+
[
|
364 |
+
'{"ops":[{"insert":"A "},{"attributes":{"color":"#FFD700"},"insert":"marble statue of a wolf\'s head and shoulder"},{"insert":", surrounded by colorful flowers michelangelo, detailed, intricate, full of color, led lighting, trending on artstation, 4 k, hyperrealistic, 3 5 mm, focused, extreme details, unreal engine 5, masterpiece "}]}',
|
365 |
+
'',
|
366 |
+
5,
|
367 |
+
0.3,
|
368 |
+
0,
|
369 |
+
5,
|
370 |
+
0.6,
|
371 |
+
None,
|
372 |
+
False
|
373 |
],
|
374 |
]
|
375 |
+
gr.Examples(examples=color_examples,
|
376 |
+
label='Font color examples',
|
|
|
377 |
inputs=[
|
378 |
text_input,
|
379 |
negative_prompt,
|
380 |
+
num_segments,
|
381 |
+
segment_threshold,
|
382 |
+
inject_interval,
|
383 |
seed,
|
384 |
color_guidance_weight,
|
385 |
rich_text_input,
|
386 |
+
background_aug,
|
387 |
],
|
388 |
outputs=[
|
389 |
plaintext_result,
|
390 |
richtext_result,
|
391 |
+
segments,
|
392 |
token_map,
|
393 |
],
|
394 |
fn=generate,
|
395 |
# cache_examples=True,
|
396 |
examples_per_page=20)
|
397 |
+
|
398 |
with gr.Row():
|
399 |
+
style_examples = [
|
400 |
[
|
401 |
+
'{"ops":[{"insert":"a "},{"attributes":{"font":"mirza"},"insert":"beautiful garden"},{"insert":" with a "},{"attributes":{"font":"roboto"},"insert":"snow mountain in the background"},{"insert":""}]}',
|
402 |
'',
|
403 |
+
5,
|
404 |
+
0.3,
|
405 |
+
0.2,
|
406 |
+
3,
|
407 |
+
0.5,
|
408 |
+
None,
|
409 |
+
False
|
410 |
],
|
411 |
[
|
412 |
+
'{"ops":[{"attributes":{"link":"the awe-inspiring sky and ocean in the style of J.M.W. Turner"},"insert":"the awe-inspiring sky and sea"},{"insert":" by "},{"attributes":{"font":"mirza"},"insert":"a coast with flowers and grasses in spring"}]}',
|
413 |
+
'worst quality, dark, poor quality',
|
414 |
+
5,
|
415 |
+
0.3,
|
416 |
+
0,
|
417 |
9,
|
418 |
+
0.5,
|
419 |
+
None,
|
420 |
+
False
|
421 |
],
|
422 |
[
|
423 |
+
'{"ops":[{"insert":"a "},{"attributes":{"font":"slabo"},"insert":"night sky filled with stars"},{"insert":" above a "},{"attributes":{"font":"roboto"},"insert":"turbulent sea with giant waves"}]}',
|
424 |
'',
|
425 |
+
2,
|
426 |
+
0.4,
|
427 |
+
0,
|
428 |
+
6,
|
429 |
+
0.5,
|
430 |
+
None,
|
431 |
+
False
|
432 |
],
|
433 |
]
|
434 |
+
gr.Examples(examples=style_examples,
|
435 |
+
label='Font style examples',
|
436 |
inputs=[
|
437 |
text_input,
|
438 |
negative_prompt,
|
439 |
+
num_segments,
|
440 |
+
segment_threshold,
|
441 |
+
inject_interval,
|
442 |
seed,
|
443 |
color_guidance_weight,
|
444 |
rich_text_input,
|
445 |
+
background_aug,
|
446 |
],
|
447 |
outputs=[
|
448 |
plaintext_result,
|
449 |
richtext_result,
|
450 |
+
segments,
|
451 |
token_map,
|
452 |
],
|
453 |
fn=generate,
|
454 |
# cache_examples=True,
|
455 |
examples_per_page=20)
|
456 |
+
|
457 |
with gr.Row():
|
458 |
size_examples = [
|
459 |
[
|
460 |
'{"ops": [{"insert": "A pizza with "}, {"attributes": {"size": "60px"}, "insert": "pineapple"}, {"insert": ", pepperoni, and mushroom on the top, 4k, photorealistic"}]}',
|
461 |
'blurry, art, painting, rendering, drawing, sketch, ugly, duplicate, morbid, mutilated, mutated, deformed, disfigured low quality, worst quality',
|
462 |
+
5,
|
463 |
+
0.3,
|
464 |
+
0,
|
465 |
13,
|
466 |
1,
|
467 |
+
None,
|
468 |
+
False
|
469 |
],
|
470 |
[
|
471 |
'{"ops": [{"insert": "A pizza with pineapple, "}, {"attributes": {"size": "20px"}, "insert": "pepperoni"}, {"insert": ", and mushroom on the top, 4k, photorealistic"}]}',
|
472 |
'blurry, art, painting, rendering, drawing, sketch, ugly, duplicate, morbid, mutilated, mutated, deformed, disfigured low quality, worst quality',
|
473 |
+
5,
|
474 |
+
0.3,
|
475 |
+
0,
|
476 |
13,
|
477 |
1,
|
478 |
+
None,
|
479 |
+
False
|
480 |
],
|
481 |
[
|
482 |
'{"ops": [{"insert": "A pizza with pineapple, pepperoni, and "}, {"attributes": {"size": "70px"}, "insert": "mushroom"}, {"insert": " on the top, 4k, photorealistic"}]}',
|
483 |
'blurry, art, painting, rendering, drawing, sketch, ugly, duplicate, morbid, mutilated, mutated, deformed, disfigured low quality, worst quality',
|
484 |
+
5,
|
485 |
+
0.3,
|
486 |
+
0,
|
487 |
13,
|
488 |
1,
|
489 |
+
None,
|
490 |
+
False
|
491 |
],
|
492 |
]
|
493 |
gr.Examples(examples=size_examples,
|
|
|
495 |
inputs=[
|
496 |
text_input,
|
497 |
negative_prompt,
|
498 |
+
num_segments,
|
499 |
+
segment_threshold,
|
500 |
+
inject_interval,
|
501 |
seed,
|
502 |
color_guidance_weight,
|
503 |
rich_text_input,
|
504 |
+
background_aug,
|
505 |
],
|
506 |
outputs=[
|
507 |
plaintext_result,
|
508 |
richtext_result,
|
509 |
+
segments,
|
510 |
token_map,
|
511 |
],
|
512 |
fn=generate,
|
|
|
521 |
width,
|
522 |
seed,
|
523 |
steps,
|
524 |
+
num_segments,
|
525 |
+
segment_threshold,
|
526 |
+
inject_interval,
|
527 |
guidance_weight,
|
528 |
color_guidance_weight,
|
529 |
+
rich_text_input,
|
530 |
+
background_aug
|
531 |
],
|
532 |
+
outputs=[plaintext_result, richtext_result, segments, token_map],
|
533 |
_js=get_js_data
|
534 |
).then(
|
535 |
fn=lambda: gr.update(visible=True), inputs=None, outputs=share_row, queue=False)
|
models/attention.py
CHANGED
@@ -492,7 +492,7 @@ class BasicTransformerBlock(nn.Module):
|
|
492 |
|
493 |
if self.only_cross_attention:
|
494 |
attn_out, _ = self.attn1(
|
495 |
-
norm_hidden_states, context, text_format_dict=text_format_dict) + hidden_states
|
496 |
hidden_states = attn_out + hidden_states
|
497 |
else:
|
498 |
attn_out, _ = self.attn1(norm_hidden_states)
|
@@ -583,7 +583,7 @@ class CrossAttention(nn.Module):
|
|
583 |
head_size, seq_len, seq_len2)
|
584 |
return tensor.mean(1)
|
585 |
|
586 |
-
def forward(self, hidden_states, context=None, mask=None, text_format_dict={}):
|
587 |
batch_size, sequence_length, _ = hidden_states.shape
|
588 |
|
589 |
query = self.to_q(hidden_states)
|
@@ -607,7 +607,7 @@ class CrossAttention(nn.Module):
|
|
607 |
if self._slice_size is None or query.shape[0] // self._slice_size == 1:
|
608 |
# only this attention function is used
|
609 |
hidden_states, attn_probs = self._attention(
|
610 |
-
query, key, value, **text_format_dict)
|
611 |
|
612 |
# linear proj
|
613 |
hidden_states = self.to_out[0](hidden_states)
|
@@ -625,11 +625,11 @@ class CrossAttention(nn.Module):
|
|
625 |
alpha=self.scale,
|
626 |
)
|
627 |
|
628 |
-
def _attention(self, query, key, value, word_pos=None, font_size=None,
|
629 |
**kwargs):
|
630 |
attention_scores = self._qk(query, key)
|
631 |
|
632 |
-
# Font size:
|
633 |
if self.is_cross_attn and word_pos is not None and font_size is not None:
|
634 |
assert key.shape[1] == 77
|
635 |
attention_score_exp = attention_scores.exp()
|
@@ -642,13 +642,25 @@ class CrossAttention(nn.Module):
|
|
642 |
else:
|
643 |
attention_probs = attention_scores.softmax(dim=-1)
|
644 |
|
645 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
646 |
|
647 |
# reshape hidden_states
|
648 |
hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
|
649 |
-
|
|
|
|
|
650 |
attention_probs)
|
651 |
-
return hidden_states, attention_probs
|
652 |
|
653 |
def _memory_efficient_attention_xformers(self, query, key, value):
|
654 |
query = query.contiguous()
|
|
|
492 |
|
493 |
if self.only_cross_attention:
|
494 |
attn_out, _ = self.attn1(
|
495 |
+
norm_hidden_states, context=context, text_format_dict=text_format_dict) + hidden_states
|
496 |
hidden_states = attn_out + hidden_states
|
497 |
else:
|
498 |
attn_out, _ = self.attn1(norm_hidden_states)
|
|
|
583 |
head_size, seq_len, seq_len2)
|
584 |
return tensor.mean(1)
|
585 |
|
586 |
+
def forward(self, hidden_states, real_attn_probs=None, context=None, mask=None, text_format_dict={}):
|
587 |
batch_size, sequence_length, _ = hidden_states.shape
|
588 |
|
589 |
query = self.to_q(hidden_states)
|
|
|
607 |
if self._slice_size is None or query.shape[0] // self._slice_size == 1:
|
608 |
# only this attention function is used
|
609 |
hidden_states, attn_probs = self._attention(
|
610 |
+
query, key, value, real_attn_probs, **text_format_dict)
|
611 |
|
612 |
# linear proj
|
613 |
hidden_states = self.to_out[0](hidden_states)
|
|
|
625 |
alpha=self.scale,
|
626 |
)
|
627 |
|
628 |
+
def _attention(self, query, key, value, real_attn_probs=None, word_pos=None, font_size=None,
|
629 |
**kwargs):
|
630 |
attention_scores = self._qk(query, key)
|
631 |
|
632 |
+
# Font size V2:
|
633 |
if self.is_cross_attn and word_pos is not None and font_size is not None:
|
634 |
assert key.shape[1] == 77
|
635 |
attention_score_exp = attention_scores.exp()
|
|
|
642 |
else:
|
643 |
attention_probs = attention_scores.softmax(dim=-1)
|
644 |
|
645 |
+
# compute attention output
|
646 |
+
if real_attn_probs is None:
|
647 |
+
hidden_states = torch.bmm(attention_probs, value)
|
648 |
+
else:
|
649 |
+
if isinstance(real_attn_probs, dict):
|
650 |
+
for pos1, pos2 in zip(real_attn_probs['inject_pos'][0], real_attn_probs['inject_pos'][1]):
|
651 |
+
attention_probs[:, :,
|
652 |
+
pos2] = real_attn_probs['reference'][:, :, pos1]
|
653 |
+
hidden_states = torch.bmm(attention_probs, value)
|
654 |
+
else:
|
655 |
+
hidden_states = torch.bmm(real_attn_probs, value)
|
656 |
|
657 |
# reshape hidden_states
|
658 |
hidden_states = self.reshape_batch_dim_to_heads(hidden_states)
|
659 |
+
|
660 |
+
# we also return the map averaged over heads to save memory footprint
|
661 |
+
attention_probs_avg = self.reshape_batch_dim_to_heads_and_average(
|
662 |
attention_probs)
|
663 |
+
return hidden_states, [attention_probs_avg, attention_probs]
|
664 |
|
665 |
def _memory_efficient_attention_xformers(self, query, key, value):
|
666 |
query = query.contiguous()
|
models/region_diffusion.py
CHANGED
@@ -6,6 +6,7 @@ from functools import partial
|
|
6 |
from transformers import CLIPTextModel, CLIPTokenizer, logging
|
7 |
from diffusers import AutoencoderKL, PNDMScheduler, EulerDiscreteScheduler, DPMSolverMultistepScheduler
|
8 |
from models.unet_2d_condition import UNet2DConditionModel
|
|
|
9 |
|
10 |
# suppress partial model loading warning
|
11 |
logging.set_verbosity_error()
|
@@ -38,6 +39,7 @@ class RegionDiffusion(nn.Module):
|
|
38 |
self.masks = []
|
39 |
self.attention_maps = None
|
40 |
self.color_loss = torch.nn.functional.mse_loss
|
|
|
41 |
|
42 |
print(f'[INFO] loaded stable diffusion!')
|
43 |
|
@@ -79,47 +81,83 @@ class RegionDiffusion(nn.Module):
|
|
79 |
return text_embeddings
|
80 |
|
81 |
def produce_latents(self, text_embeddings, height=512, width=512, num_inference_steps=50, guidance_scale=7.5,
|
82 |
-
latents=None,
|
83 |
|
84 |
if latents is None:
|
85 |
latents = torch.randn(
|
86 |
(1, self.unet.in_channels, height // 8, width // 8), device=self.device)
|
87 |
|
|
|
|
|
88 |
self.scheduler.set_timesteps(num_inference_steps)
|
89 |
n_styles = text_embeddings.shape[0]-1
|
90 |
assert n_styles == len(self.masks)
|
91 |
-
|
92 |
with torch.autocast('cuda'):
|
93 |
for i, t in enumerate(self.scheduler.timesteps):
|
94 |
|
95 |
# predict the noise residual
|
96 |
with torch.no_grad():
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
|
|
104 |
text_format_dict={})['sample']
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
105 |
else:
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
|
|
112 |
|
113 |
# perform classifier-free guidance
|
114 |
noise_pred = noise_pred_uncond + guidance_scale * \
|
115 |
(noise_pred_text - noise_pred_uncond)
|
116 |
|
117 |
-
|
118 |
-
|
119 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
120 |
|
121 |
-
# apply
|
122 |
-
if
|
123 |
with torch.enable_grad():
|
124 |
if not latents.requires_grad:
|
125 |
latents.requires_grad = True
|
@@ -137,7 +175,7 @@ class RegionDiffusion(nn.Module):
|
|
137 |
loss_total += loss
|
138 |
loss_total.backward()
|
139 |
latents = (
|
140 |
-
latents - latents.grad * text_format_dict['color_guidance_weight']).detach().clone()
|
141 |
|
142 |
return latents
|
143 |
|
@@ -162,6 +200,7 @@ class RegionDiffusion(nn.Module):
|
|
162 |
(text_embeddings.shape[0] // 2, self.unet.in_channels, height // 8, width // 8), device=self.device)
|
163 |
|
164 |
self.scheduler.set_timesteps(num_inference_steps)
|
|
|
165 |
|
166 |
with torch.autocast('cuda'):
|
167 |
for i, t in enumerate(self.scheduler.timesteps):
|
@@ -202,8 +241,18 @@ class RegionDiffusion(nn.Module):
|
|
202 |
|
203 |
return imgs
|
204 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
205 |
def prompt_to_img(self, prompts, negative_prompts='', height=512, width=512, num_inference_steps=50,
|
206 |
-
guidance_scale=7.5, latents=None, text_format_dict={},
|
207 |
|
208 |
if isinstance(prompts, str):
|
209 |
prompts = [prompts]
|
@@ -215,18 +264,11 @@ class RegionDiffusion(nn.Module):
|
|
215 |
text_embeds = self.get_text_embeds(
|
216 |
prompts, negative_prompts) # [2, 77, 768]
|
217 |
|
218 |
-
if len(text_format_dict) > 0:
|
219 |
-
if 'font_styles' in text_format_dict and text_format_dict['font_styles'] is not None:
|
220 |
-
text_format_dict['font_styles_embs'] = self.get_text_embeds_list(
|
221 |
-
text_format_dict['font_styles']) # [2, 77, 768]
|
222 |
-
else:
|
223 |
-
text_format_dict['font_styles_embs'] = None
|
224 |
-
|
225 |
# else:
|
226 |
latents = self.produce_latents(text_embeds, height=height, width=width, latents=latents,
|
227 |
num_inference_steps=num_inference_steps, guidance_scale=guidance_scale,
|
228 |
-
|
229 |
-
|
230 |
# Img latents -> imgs
|
231 |
imgs = self.decode_latents(latents) # [1, 3, 512, 512]
|
232 |
|
@@ -272,7 +314,156 @@ class RegionDiffusion(nn.Module):
|
|
272 |
# attention_dict is a dictionary containing attention maps for every attention layer
|
273 |
self.attention_maps = attention_dict
|
274 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
275 |
def remove_evaluation_hooks(self):
|
276 |
for hook in self.forward_hooks:
|
277 |
hook.remove()
|
278 |
self.attention_maps = None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
from transformers import CLIPTextModel, CLIPTokenizer, logging
|
7 |
from diffusers import AutoencoderKL, PNDMScheduler, EulerDiscreteScheduler, DPMSolverMultistepScheduler
|
8 |
from models.unet_2d_condition import UNet2DConditionModel
|
9 |
+
from utils.attention_utils import CrossAttentionLayers, SelfAttentionLayers
|
10 |
|
11 |
# suppress partial model loading warning
|
12 |
logging.set_verbosity_error()
|
|
|
39 |
self.masks = []
|
40 |
self.attention_maps = None
|
41 |
self.color_loss = torch.nn.functional.mse_loss
|
42 |
+
self.forward_replacement_hooks = []
|
43 |
|
44 |
print(f'[INFO] loaded stable diffusion!')
|
45 |
|
|
|
81 |
return text_embeddings
|
82 |
|
83 |
def produce_latents(self, text_embeddings, height=512, width=512, num_inference_steps=50, guidance_scale=7.5,
|
84 |
+
latents=None, use_guidance=False, text_format_dict={}, inject_selfattn=0, bg_aug_end=1000):
|
85 |
|
86 |
if latents is None:
|
87 |
latents = torch.randn(
|
88 |
(1, self.unet.in_channels, height // 8, width // 8), device=self.device)
|
89 |
|
90 |
+
if inject_selfattn > 0:
|
91 |
+
latents_reference = latents.clone().detach()
|
92 |
self.scheduler.set_timesteps(num_inference_steps)
|
93 |
n_styles = text_embeddings.shape[0]-1
|
94 |
assert n_styles == len(self.masks)
|
|
|
95 |
with torch.autocast('cuda'):
|
96 |
for i, t in enumerate(self.scheduler.timesteps):
|
97 |
|
98 |
# predict the noise residual
|
99 |
with torch.no_grad():
|
100 |
+
# tokens without any attributes
|
101 |
+
feat_inject_step = t > (1-inject_selfattn) * 1000
|
102 |
+
noise_pred_uncond_cur = self.unet(latents, t, encoder_hidden_states=text_embeddings[:1],
|
103 |
+
text_format_dict={})['sample']
|
104 |
+
noise_pred_text_cur = self.unet(latents, t, encoder_hidden_states=text_embeddings[-1:],
|
105 |
+
text_format_dict=text_format_dict)['sample']
|
106 |
+
if inject_selfattn > 0:
|
107 |
+
noise_pred_uncond_refer = self.unet(latents_reference, t, encoder_hidden_states=text_embeddings[:1],
|
108 |
text_format_dict={})['sample']
|
109 |
+
self.register_selfattn_hooks(feat_inject_step)
|
110 |
+
noise_pred_text_refer = self.unet(latents_reference, t, encoder_hidden_states=text_embeddings[-1:],
|
111 |
+
text_format_dict={})['sample']
|
112 |
+
self.remove_selfattn_hooks()
|
113 |
+
noise_pred_uncond = noise_pred_uncond_cur * self.masks[-1]
|
114 |
+
noise_pred_text = noise_pred_text_cur * self.masks[-1]
|
115 |
+
# tokens with attributes
|
116 |
+
for style_i, mask in enumerate(self.masks[:-1]):
|
117 |
+
if t > bg_aug_end:
|
118 |
+
rand_rgb = torch.rand([1, 3, 1, 1]).cuda()
|
119 |
+
black_background = torch.ones(
|
120 |
+
[1, 3, height, width]).cuda()*rand_rgb
|
121 |
+
black_latent = self.encode_imgs(
|
122 |
+
black_background)
|
123 |
+
noise = torch.randn_like(black_latent)
|
124 |
+
black_latent_noisy = self.scheduler.add_noise(
|
125 |
+
black_latent, noise, t)
|
126 |
+
masked_latent = (
|
127 |
+
mask > 0.001) * latents + (mask < 0.001) * black_latent_noisy
|
128 |
+
noise_pred_uncond_cur = self.unet(masked_latent, t, encoder_hidden_states=text_embeddings[:1],
|
129 |
+
text_format_dict={})['sample']
|
130 |
else:
|
131 |
+
masked_latent = latents
|
132 |
+
self.register_replacement_hooks(feat_inject_step)
|
133 |
+
noise_pred_text_cur = self.unet(masked_latent, t, encoder_hidden_states=text_embeddings[style_i+1:style_i+2],
|
134 |
+
text_format_dict={})['sample']
|
135 |
+
self.remove_replacement_hooks()
|
136 |
+
noise_pred_uncond = noise_pred_uncond + noise_pred_uncond_cur*mask
|
137 |
+
noise_pred_text = noise_pred_text + noise_pred_text_cur*mask
|
138 |
|
139 |
# perform classifier-free guidance
|
140 |
noise_pred = noise_pred_uncond + guidance_scale * \
|
141 |
(noise_pred_text - noise_pred_uncond)
|
142 |
|
143 |
+
if inject_selfattn > 0:
|
144 |
+
noise_pred_refer = noise_pred_uncond_refer + guidance_scale * \
|
145 |
+
(noise_pred_text_refer - noise_pred_uncond_refer)
|
146 |
+
|
147 |
+
# compute the previous noisy sample x_t -> x_t-1
|
148 |
+
latents_reference = self.scheduler.step(torch.cat([noise_pred, noise_pred_refer]), t,
|
149 |
+
torch.cat([latents, latents_reference]))[
|
150 |
+
'prev_sample']
|
151 |
+
latents, latents_reference = torch.chunk(
|
152 |
+
latents_reference, 2, dim=0)
|
153 |
+
|
154 |
+
else:
|
155 |
+
# compute the previous noisy sample x_t -> x_t-1
|
156 |
+
latents = self.scheduler.step(noise_pred, t, latents)[
|
157 |
+
'prev_sample']
|
158 |
|
159 |
+
# apply guidance
|
160 |
+
if use_guidance and t < text_format_dict['guidance_start_step']:
|
161 |
with torch.enable_grad():
|
162 |
if not latents.requires_grad:
|
163 |
latents.requires_grad = True
|
|
|
175 |
loss_total += loss
|
176 |
loss_total.backward()
|
177 |
latents = (
|
178 |
+
latents - latents.grad * text_format_dict['color_guidance_weight'] * self.masks[0]).detach().clone()
|
179 |
|
180 |
return latents
|
181 |
|
|
|
200 |
(text_embeddings.shape[0] // 2, self.unet.in_channels, height // 8, width // 8), device=self.device)
|
201 |
|
202 |
self.scheduler.set_timesteps(num_inference_steps)
|
203 |
+
self.remove_replacement_hooks()
|
204 |
|
205 |
with torch.autocast('cuda'):
|
206 |
for i, t in enumerate(self.scheduler.timesteps):
|
|
|
241 |
|
242 |
return imgs
|
243 |
|
244 |
+
def encode_imgs(self, imgs):
|
245 |
+
# imgs: [B, 3, H, W]
|
246 |
+
|
247 |
+
imgs = 2 * imgs - 1
|
248 |
+
|
249 |
+
posterior = self.vae.encode(imgs).latent_dist
|
250 |
+
latents = posterior.sample() * 0.18215
|
251 |
+
|
252 |
+
return latents
|
253 |
+
|
254 |
def prompt_to_img(self, prompts, negative_prompts='', height=512, width=512, num_inference_steps=50,
|
255 |
+
guidance_scale=7.5, latents=None, text_format_dict={}, use_guidance=False, inject_selfattn=0, bg_aug_end=1000):
|
256 |
|
257 |
if isinstance(prompts, str):
|
258 |
prompts = [prompts]
|
|
|
264 |
text_embeds = self.get_text_embeds(
|
265 |
prompts, negative_prompts) # [2, 77, 768]
|
266 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
267 |
# else:
|
268 |
latents = self.produce_latents(text_embeds, height=height, width=width, latents=latents,
|
269 |
num_inference_steps=num_inference_steps, guidance_scale=guidance_scale,
|
270 |
+
use_guidance=use_guidance, text_format_dict=text_format_dict,
|
271 |
+
inject_selfattn=inject_selfattn, bg_aug_end=bg_aug_end) # [1, 4, 64, 64]
|
272 |
# Img latents -> imgs
|
273 |
imgs = self.decode_latents(latents) # [1, 3, 512, 512]
|
274 |
|
|
|
314 |
# attention_dict is a dictionary containing attention maps for every attention layer
|
315 |
self.attention_maps = attention_dict
|
316 |
|
317 |
+
def register_selfattn_hooks(self, feat_inject_step=False):
|
318 |
+
r"""Function for registering hooks during evaluation.
|
319 |
+
We mainly store activation maps averaged over queries.
|
320 |
+
"""
|
321 |
+
self.selfattn_forward_hooks = []
|
322 |
+
|
323 |
+
def save_activations(activations, name, module, inp, out):
|
324 |
+
r"""
|
325 |
+
PyTorch Forward hook to save outputs at each forward pass.
|
326 |
+
"""
|
327 |
+
# out[0] - final output of attention layer
|
328 |
+
# out[1] - attention probability matrix
|
329 |
+
if 'attn2' in name:
|
330 |
+
assert out[1][1].shape[-1] == 77
|
331 |
+
# cross attention injection
|
332 |
+
# activations[name] = out[1][1].detach()
|
333 |
+
else:
|
334 |
+
assert out[1][1].shape[-1] != 77
|
335 |
+
activations[name] = out[1][1].detach()
|
336 |
+
|
337 |
+
def save_resnet_activations(activations, name, module, inp, out):
|
338 |
+
r"""
|
339 |
+
PyTorch Forward hook to save outputs at each forward pass.
|
340 |
+
"""
|
341 |
+
# out[0] - final output of residual layer
|
342 |
+
# out[1] - residual hidden feature
|
343 |
+
# import ipdb
|
344 |
+
# ipdb.set_trace()
|
345 |
+
assert out[1].shape[-1] == 16
|
346 |
+
activations[name] = out[1].detach()
|
347 |
+
attention_dict = collections.defaultdict(list)
|
348 |
+
for name, module in self.unet.named_modules():
|
349 |
+
leaf_name = name.split('.')[-1]
|
350 |
+
if 'attn' in leaf_name and feat_inject_step:
|
351 |
+
# Register hook to obtain outputs at every attention layer.
|
352 |
+
self.selfattn_forward_hooks.append(module.register_forward_hook(
|
353 |
+
partial(save_activations, attention_dict, name)
|
354 |
+
))
|
355 |
+
if name == 'up_blocks.1.resnets.1' and feat_inject_step:
|
356 |
+
self.selfattn_forward_hooks.append(module.register_forward_hook(
|
357 |
+
partial(save_resnet_activations, attention_dict, name)
|
358 |
+
))
|
359 |
+
# attention_dict is a dictionary containing attention maps for every attention layer
|
360 |
+
self.self_attention_maps_cur = attention_dict
|
361 |
+
|
362 |
+
def register_replacement_hooks(self, feat_inject_step=False):
|
363 |
+
r"""Function for registering hooks to replace self attention.
|
364 |
+
"""
|
365 |
+
self.forward_replacement_hooks = []
|
366 |
+
|
367 |
+
def replace_activations(name, module, args):
|
368 |
+
r"""
|
369 |
+
PyTorch Forward hook to save outputs at each forward pass.
|
370 |
+
"""
|
371 |
+
if 'attn1' in name:
|
372 |
+
modified_args = (args[0], self.self_attention_maps_cur[name])
|
373 |
+
return modified_args
|
374 |
+
# cross attention injection
|
375 |
+
# elif 'attn2' in name:
|
376 |
+
# modified_map = {
|
377 |
+
# 'reference': self.self_attention_maps_cur[name],
|
378 |
+
# 'inject_pos': self.inject_pos,
|
379 |
+
# }
|
380 |
+
# modified_args = (args[0], modified_map)
|
381 |
+
# return modified_args
|
382 |
+
|
383 |
+
def replace_resnet_activations(name, module, args):
|
384 |
+
r"""
|
385 |
+
PyTorch Forward hook to save outputs at each forward pass.
|
386 |
+
"""
|
387 |
+
modified_args = (args[0], args[1],
|
388 |
+
self.self_attention_maps_cur[name])
|
389 |
+
return modified_args
|
390 |
+
for name, module in self.unet.named_modules():
|
391 |
+
leaf_name = name.split('.')[-1]
|
392 |
+
if 'attn' in leaf_name and feat_inject_step:
|
393 |
+
# Register hook to obtain outputs at every attention layer.
|
394 |
+
self.forward_replacement_hooks.append(module.register_forward_pre_hook(
|
395 |
+
partial(replace_activations, name)
|
396 |
+
))
|
397 |
+
if name == 'up_blocks.1.resnets.1' and feat_inject_step:
|
398 |
+
# Register hook to obtain outputs at every attention layer.
|
399 |
+
self.forward_replacement_hooks.append(module.register_forward_pre_hook(
|
400 |
+
partial(replace_resnet_activations, name)
|
401 |
+
))
|
402 |
+
|
403 |
+
def register_tokenmap_hooks(self):
|
404 |
+
r"""Function for registering hooks during evaluation.
|
405 |
+
We mainly store activation maps averaged over queries.
|
406 |
+
"""
|
407 |
+
self.forward_hooks = []
|
408 |
+
|
409 |
+
def save_activations(selfattn_maps, crossattn_maps, n_maps, name, module, inp, out):
|
410 |
+
r"""
|
411 |
+
PyTorch Forward hook to save outputs at each forward pass.
|
412 |
+
"""
|
413 |
+
# out[0] - final output of attention layer
|
414 |
+
# out[1] - attention probability matrices
|
415 |
+
if name in n_maps:
|
416 |
+
n_maps[name] += 1
|
417 |
+
else:
|
418 |
+
n_maps[name] = 1
|
419 |
+
if 'attn2' in name:
|
420 |
+
assert out[1][0].shape[-1] == 77
|
421 |
+
if name in CrossAttentionLayers and n_maps[name] > 10:
|
422 |
+
if name in crossattn_maps:
|
423 |
+
crossattn_maps[name] += out[1][0].detach().cpu()[1:2]
|
424 |
+
else:
|
425 |
+
crossattn_maps[name] = out[1][0].detach().cpu()[1:2]
|
426 |
+
else:
|
427 |
+
assert out[1][0].shape[-1] != 77
|
428 |
+
if name in SelfAttentionLayers and n_maps[name] > 10:
|
429 |
+
if name in crossattn_maps:
|
430 |
+
selfattn_maps[name] += out[1][0].detach().cpu()[1:2]
|
431 |
+
else:
|
432 |
+
selfattn_maps[name] = out[1][0].detach().cpu()[1:2]
|
433 |
+
|
434 |
+
selfattn_maps = collections.defaultdict(list)
|
435 |
+
crossattn_maps = collections.defaultdict(list)
|
436 |
+
n_maps = collections.defaultdict(list)
|
437 |
+
|
438 |
+
for name, module in self.unet.named_modules():
|
439 |
+
leaf_name = name.split('.')[-1]
|
440 |
+
if 'attn' in leaf_name:
|
441 |
+
# Register hook to obtain outputs at every attention layer.
|
442 |
+
self.forward_hooks.append(module.register_forward_hook(
|
443 |
+
partial(save_activations, selfattn_maps,
|
444 |
+
crossattn_maps, n_maps, name)
|
445 |
+
))
|
446 |
+
# attention_dict is a dictionary containing attention maps for every attention layer
|
447 |
+
self.selfattn_maps = selfattn_maps
|
448 |
+
self.crossattn_maps = crossattn_maps
|
449 |
+
self.n_maps = n_maps
|
450 |
+
|
451 |
+
def remove_tokenmap_hooks(self):
|
452 |
+
for hook in self.forward_hooks:
|
453 |
+
hook.remove()
|
454 |
+
self.selfattn_maps = None
|
455 |
+
self.crossattn_maps = None
|
456 |
+
self.n_maps = None
|
457 |
+
|
458 |
def remove_evaluation_hooks(self):
|
459 |
for hook in self.forward_hooks:
|
460 |
hook.remove()
|
461 |
self.attention_maps = None
|
462 |
+
|
463 |
+
def remove_replacement_hooks(self):
|
464 |
+
for hook in self.forward_replacement_hooks:
|
465 |
+
hook.remove()
|
466 |
+
|
467 |
+
def remove_selfattn_hooks(self):
|
468 |
+
for hook in self.selfattn_forward_hooks:
|
469 |
+
hook.remove()
|
models/unet_2d_blocks.py
CHANGED
@@ -16,7 +16,7 @@ import torch
|
|
16 |
from torch import nn
|
17 |
|
18 |
from .attention import AttentionBlock, DualTransformer2DModel, Transformer2DModel
|
19 |
-
from diffusers.models.resnet import Downsample2D, FirDownsample2D, FirUpsample2D,
|
20 |
|
21 |
|
22 |
def get_down_block(
|
@@ -36,7 +36,8 @@ def get_down_block(
|
|
36 |
use_linear_projection=False,
|
37 |
only_cross_attention=False,
|
38 |
):
|
39 |
-
down_block_type = down_block_type[7:] if down_block_type.startswith(
|
|
|
40 |
if down_block_type == "DownBlock2D":
|
41 |
return DownBlock2D(
|
42 |
num_layers=num_layers,
|
@@ -64,7 +65,8 @@ def get_down_block(
|
|
64 |
)
|
65 |
elif down_block_type == "CrossAttnDownBlock2D":
|
66 |
if cross_attention_dim is None:
|
67 |
-
raise ValueError(
|
|
|
68 |
return CrossAttnDownBlock2D(
|
69 |
num_layers=num_layers,
|
70 |
in_channels=in_channels,
|
@@ -147,7 +149,8 @@ def get_up_block(
|
|
147 |
use_linear_projection=False,
|
148 |
only_cross_attention=False,
|
149 |
):
|
150 |
-
up_block_type = up_block_type[7:] if up_block_type.startswith(
|
|
|
151 |
if up_block_type == "UpBlock2D":
|
152 |
return UpBlock2D(
|
153 |
num_layers=num_layers,
|
@@ -162,7 +165,8 @@ def get_up_block(
|
|
162 |
)
|
163 |
elif up_block_type == "CrossAttnUpBlock2D":
|
164 |
if cross_attention_dim is None:
|
165 |
-
raise ValueError(
|
|
|
166 |
return CrossAttnUpBlock2D(
|
167 |
num_layers=num_layers,
|
168 |
in_channels=in_channels,
|
@@ -258,7 +262,8 @@ class UNetMidBlock2D(nn.Module):
|
|
258 |
super().__init__()
|
259 |
|
260 |
self.attention_type = attention_type
|
261 |
-
resnet_groups = resnet_groups if resnet_groups is not None else min(
|
|
|
262 |
|
263 |
# there is always at least one resnet
|
264 |
resnets = [
|
@@ -312,7 +317,7 @@ class UNetMidBlock2D(nn.Module):
|
|
312 |
hidden_states = attn(hidden_states)
|
313 |
else:
|
314 |
hidden_states = attn(hidden_states, encoder_states)
|
315 |
-
hidden_states = resnet(hidden_states, temb)
|
316 |
|
317 |
return hidden_states
|
318 |
|
@@ -340,7 +345,8 @@ class UNetMidBlock2DCrossAttn(nn.Module):
|
|
340 |
|
341 |
self.attention_type = attention_type
|
342 |
self.attn_num_head_channels = attn_num_head_channels
|
343 |
-
resnet_groups = resnet_groups if resnet_groups is not None else min(
|
|
|
344 |
|
345 |
# there is always at least one resnet
|
346 |
resnets = [
|
@@ -420,15 +426,16 @@ class UNetMidBlock2DCrossAttn(nn.Module):
|
|
420 |
|
421 |
def set_use_memory_efficient_attention_xformers(self, use_memory_efficient_attention_xformers: bool):
|
422 |
for attn in self.attentions:
|
423 |
-
attn._set_use_memory_efficient_attention_xformers(
|
|
|
424 |
|
425 |
def forward(self, hidden_states, temb=None, encoder_hidden_states=None,
|
426 |
text_format_dict={}):
|
427 |
-
hidden_states = self.resnets[0](hidden_states, temb)
|
428 |
for attn, resnet in zip(self.attentions, self.resnets[1:]):
|
429 |
-
hidden_states = attn(hidden_states, encoder_hidden_states,
|
430 |
text_format_dict).sample
|
431 |
-
hidden_states = resnet(hidden_states, temb)
|
432 |
|
433 |
return hidden_states
|
434 |
|
@@ -502,7 +509,7 @@ class AttnDownBlock2D(nn.Module):
|
|
502 |
output_states = ()
|
503 |
|
504 |
for resnet, attn in zip(self.resnets, self.attentions):
|
505 |
-
hidden_states = resnet(hidden_states, temb)
|
506 |
hidden_states = attn(hidden_states)
|
507 |
output_states += (hidden_states,)
|
508 |
|
@@ -620,7 +627,8 @@ class CrossAttnDownBlock2D(nn.Module):
|
|
620 |
|
621 |
def set_use_memory_efficient_attention_xformers(self, use_memory_efficient_attention_xformers: bool):
|
622 |
for attn in self.attentions:
|
623 |
-
attn._set_use_memory_efficient_attention_xformers(
|
|
|
624 |
|
625 |
def forward(self, hidden_states, temb=None, encoder_hidden_states=None,
|
626 |
text_format_dict={}):
|
@@ -638,13 +646,15 @@ class CrossAttnDownBlock2D(nn.Module):
|
|
638 |
|
639 |
return custom_forward
|
640 |
|
641 |
-
hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
|
642 |
hidden_states = torch.utils.checkpoint.checkpoint(
|
643 |
-
create_custom_forward(
|
644 |
-
|
|
|
|
|
|
|
645 |
)[0]
|
646 |
else:
|
647 |
-
hidden_states = resnet(hidden_states, temb)
|
648 |
hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states,
|
649 |
text_format_dict=text_format_dict).sample
|
650 |
|
@@ -723,9 +733,10 @@ class DownBlock2D(nn.Module):
|
|
723 |
|
724 |
return custom_forward
|
725 |
|
726 |
-
hidden_states = torch.utils.checkpoint.checkpoint(
|
|
|
727 |
else:
|
728 |
-
hidden_states = resnet(hidden_states, temb)
|
729 |
|
730 |
output_states += (hidden_states,)
|
731 |
|
@@ -789,7 +800,7 @@ class DownEncoderBlock2D(nn.Module):
|
|
789 |
|
790 |
def forward(self, hidden_states):
|
791 |
for resnet in self.resnets:
|
792 |
-
hidden_states = resnet(hidden_states, temb=None)
|
793 |
|
794 |
if self.downsamplers is not None:
|
795 |
for downsampler in self.downsamplers:
|
@@ -861,7 +872,7 @@ class AttnDownEncoderBlock2D(nn.Module):
|
|
861 |
|
862 |
def forward(self, hidden_states):
|
863 |
for resnet, attn in zip(self.resnets, self.attentions):
|
864 |
-
hidden_states = resnet(hidden_states, temb=None)
|
865 |
hidden_states = attn(hidden_states)
|
866 |
|
867 |
if self.downsamplers is not None:
|
@@ -937,8 +948,10 @@ class AttnSkipDownBlock2D(nn.Module):
|
|
937 |
down=True,
|
938 |
kernel="fir",
|
939 |
)
|
940 |
-
self.downsamplers = nn.ModuleList(
|
941 |
-
|
|
|
|
|
942 |
else:
|
943 |
self.resnet_down = None
|
944 |
self.downsamplers = None
|
@@ -948,7 +961,7 @@ class AttnSkipDownBlock2D(nn.Module):
|
|
948 |
output_states = ()
|
949 |
|
950 |
for resnet, attn in zip(self.resnets, self.attentions):
|
951 |
-
hidden_states = resnet(hidden_states, temb)
|
952 |
hidden_states = attn(hidden_states)
|
953 |
output_states += (hidden_states,)
|
954 |
|
@@ -1017,8 +1030,10 @@ class SkipDownBlock2D(nn.Module):
|
|
1017 |
down=True,
|
1018 |
kernel="fir",
|
1019 |
)
|
1020 |
-
self.downsamplers = nn.ModuleList(
|
1021 |
-
|
|
|
|
|
1022 |
else:
|
1023 |
self.resnet_down = None
|
1024 |
self.downsamplers = None
|
@@ -1028,7 +1043,7 @@ class SkipDownBlock2D(nn.Module):
|
|
1028 |
output_states = ()
|
1029 |
|
1030 |
for resnet in self.resnets:
|
1031 |
-
hidden_states = resnet(hidden_states, temb)
|
1032 |
output_states += (hidden_states,)
|
1033 |
|
1034 |
if self.downsamplers is not None:
|
@@ -1069,7 +1084,8 @@ class AttnUpBlock2D(nn.Module):
|
|
1069 |
self.attention_type = attention_type
|
1070 |
|
1071 |
for i in range(num_layers):
|
1072 |
-
res_skip_channels = in_channels if (
|
|
|
1073 |
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
1074 |
|
1075 |
resnets.append(
|
@@ -1100,7 +1116,8 @@ class AttnUpBlock2D(nn.Module):
|
|
1100 |
self.resnets = nn.ModuleList(resnets)
|
1101 |
|
1102 |
if add_upsample:
|
1103 |
-
self.upsamplers = nn.ModuleList(
|
|
|
1104 |
else:
|
1105 |
self.upsamplers = None
|
1106 |
|
@@ -1109,9 +1126,10 @@ class AttnUpBlock2D(nn.Module):
|
|
1109 |
# pop res hidden states
|
1110 |
res_hidden_states = res_hidden_states_tuple[-1]
|
1111 |
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
1112 |
-
hidden_states = torch.cat(
|
|
|
1113 |
|
1114 |
-
hidden_states = resnet(hidden_states, temb)
|
1115 |
hidden_states = attn(hidden_states)
|
1116 |
|
1117 |
if self.upsamplers is not None:
|
@@ -1152,7 +1170,8 @@ class CrossAttnUpBlock2D(nn.Module):
|
|
1152 |
self.attn_num_head_channels = attn_num_head_channels
|
1153 |
|
1154 |
for i in range(num_layers):
|
1155 |
-
res_skip_channels = in_channels if (
|
|
|
1156 |
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
1157 |
|
1158 |
resnets.append(
|
@@ -1197,7 +1216,8 @@ class CrossAttnUpBlock2D(nn.Module):
|
|
1197 |
self.resnets = nn.ModuleList(resnets)
|
1198 |
|
1199 |
if add_upsample:
|
1200 |
-
self.upsamplers = nn.ModuleList(
|
|
|
1201 |
else:
|
1202 |
self.upsamplers = None
|
1203 |
|
@@ -1224,7 +1244,8 @@ class CrossAttnUpBlock2D(nn.Module):
|
|
1224 |
|
1225 |
def set_use_memory_efficient_attention_xformers(self, use_memory_efficient_attention_xformers: bool):
|
1226 |
for attn in self.attentions:
|
1227 |
-
attn._set_use_memory_efficient_attention_xformers(
|
|
|
1228 |
|
1229 |
def forward(
|
1230 |
self,
|
@@ -1239,7 +1260,8 @@ class CrossAttnUpBlock2D(nn.Module):
|
|
1239 |
# pop res hidden states
|
1240 |
res_hidden_states = res_hidden_states_tuple[-1]
|
1241 |
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
1242 |
-
hidden_states = torch.cat(
|
|
|
1243 |
|
1244 |
if self.training and self.gradient_checkpointing:
|
1245 |
|
@@ -1252,13 +1274,15 @@ class CrossAttnUpBlock2D(nn.Module):
|
|
1252 |
|
1253 |
return custom_forward
|
1254 |
|
1255 |
-
hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
|
1256 |
hidden_states = torch.utils.checkpoint.checkpoint(
|
1257 |
-
create_custom_forward(
|
1258 |
-
|
|
|
|
|
|
|
1259 |
)[0]
|
1260 |
else:
|
1261 |
-
hidden_states = resnet(hidden_states, temb)
|
1262 |
hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states,
|
1263 |
text_format_dict=text_format_dict).sample
|
1264 |
|
@@ -1290,7 +1314,8 @@ class UpBlock2D(nn.Module):
|
|
1290 |
resnets = []
|
1291 |
|
1292 |
for i in range(num_layers):
|
1293 |
-
res_skip_channels = in_channels if (
|
|
|
1294 |
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
1295 |
|
1296 |
resnets.append(
|
@@ -1311,7 +1336,8 @@ class UpBlock2D(nn.Module):
|
|
1311 |
self.resnets = nn.ModuleList(resnets)
|
1312 |
|
1313 |
if add_upsample:
|
1314 |
-
self.upsamplers = nn.ModuleList(
|
|
|
1315 |
else:
|
1316 |
self.upsamplers = None
|
1317 |
|
@@ -1322,7 +1348,8 @@ class UpBlock2D(nn.Module):
|
|
1322 |
# pop res hidden states
|
1323 |
res_hidden_states = res_hidden_states_tuple[-1]
|
1324 |
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
1325 |
-
hidden_states = torch.cat(
|
|
|
1326 |
|
1327 |
if self.training and self.gradient_checkpointing:
|
1328 |
|
@@ -1332,9 +1359,10 @@ class UpBlock2D(nn.Module):
|
|
1332 |
|
1333 |
return custom_forward
|
1334 |
|
1335 |
-
hidden_states = torch.utils.checkpoint.checkpoint(
|
|
|
1336 |
else:
|
1337 |
-
hidden_states = resnet(hidden_states, temb)
|
1338 |
|
1339 |
if self.upsamplers is not None:
|
1340 |
for upsampler in self.upsamplers:
|
@@ -1382,13 +1410,14 @@ class UpDecoderBlock2D(nn.Module):
|
|
1382 |
self.resnets = nn.ModuleList(resnets)
|
1383 |
|
1384 |
if add_upsample:
|
1385 |
-
self.upsamplers = nn.ModuleList(
|
|
|
1386 |
else:
|
1387 |
self.upsamplers = None
|
1388 |
|
1389 |
def forward(self, hidden_states):
|
1390 |
for resnet in self.resnets:
|
1391 |
-
hidden_states = resnet(hidden_states, temb=None)
|
1392 |
|
1393 |
if self.upsamplers is not None:
|
1394 |
for upsampler in self.upsamplers:
|
@@ -1448,13 +1477,14 @@ class AttnUpDecoderBlock2D(nn.Module):
|
|
1448 |
self.resnets = nn.ModuleList(resnets)
|
1449 |
|
1450 |
if add_upsample:
|
1451 |
-
self.upsamplers = nn.ModuleList(
|
|
|
1452 |
else:
|
1453 |
self.upsamplers = None
|
1454 |
|
1455 |
def forward(self, hidden_states):
|
1456 |
for resnet, attn in zip(self.resnets, self.attentions):
|
1457 |
-
hidden_states = resnet(hidden_states, temb=None)
|
1458 |
hidden_states = attn(hidden_states)
|
1459 |
|
1460 |
if self.upsamplers is not None:
|
@@ -1490,7 +1520,8 @@ class AttnSkipUpBlock2D(nn.Module):
|
|
1490 |
self.attention_type = attention_type
|
1491 |
|
1492 |
for i in range(num_layers):
|
1493 |
-
res_skip_channels = in_channels if (
|
|
|
1494 |
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
1495 |
|
1496 |
self.resnets.append(
|
@@ -1499,7 +1530,8 @@ class AttnSkipUpBlock2D(nn.Module):
|
|
1499 |
out_channels=out_channels,
|
1500 |
temb_channels=temb_channels,
|
1501 |
eps=resnet_eps,
|
1502 |
-
groups=min(resnet_in_channels +
|
|
|
1503 |
groups_out=min(out_channels // 4, 32),
|
1504 |
dropout=dropout,
|
1505 |
time_embedding_norm=resnet_time_scale_shift,
|
@@ -1536,7 +1568,8 @@ class AttnSkipUpBlock2D(nn.Module):
|
|
1536 |
up=True,
|
1537 |
kernel="fir",
|
1538 |
)
|
1539 |
-
self.skip_conv = nn.Conv2d(out_channels, 3, kernel_size=(
|
|
|
1540 |
self.skip_norm = torch.nn.GroupNorm(
|
1541 |
num_groups=min(out_channels // 4, 32), num_channels=out_channels, eps=resnet_eps, affine=True
|
1542 |
)
|
@@ -1552,9 +1585,10 @@ class AttnSkipUpBlock2D(nn.Module):
|
|
1552 |
# pop res hidden states
|
1553 |
res_hidden_states = res_hidden_states_tuple[-1]
|
1554 |
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
1555 |
-
hidden_states = torch.cat(
|
|
|
1556 |
|
1557 |
-
hidden_states = resnet(hidden_states, temb)
|
1558 |
|
1559 |
hidden_states = self.attentions[0](hidden_states)
|
1560 |
|
@@ -1596,7 +1630,8 @@ class SkipUpBlock2D(nn.Module):
|
|
1596 |
self.resnets = nn.ModuleList([])
|
1597 |
|
1598 |
for i in range(num_layers):
|
1599 |
-
res_skip_channels = in_channels if (
|
|
|
1600 |
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
1601 |
|
1602 |
self.resnets.append(
|
@@ -1605,7 +1640,8 @@ class SkipUpBlock2D(nn.Module):
|
|
1605 |
out_channels=out_channels,
|
1606 |
temb_channels=temb_channels,
|
1607 |
eps=resnet_eps,
|
1608 |
-
groups=min(
|
|
|
1609 |
groups_out=min(out_channels // 4, 32),
|
1610 |
dropout=dropout,
|
1611 |
time_embedding_norm=resnet_time_scale_shift,
|
@@ -1633,7 +1669,8 @@ class SkipUpBlock2D(nn.Module):
|
|
1633 |
up=True,
|
1634 |
kernel="fir",
|
1635 |
)
|
1636 |
-
self.skip_conv = nn.Conv2d(out_channels, 3, kernel_size=(
|
|
|
1637 |
self.skip_norm = torch.nn.GroupNorm(
|
1638 |
num_groups=min(out_channels // 4, 32), num_channels=out_channels, eps=resnet_eps, affine=True
|
1639 |
)
|
@@ -1649,9 +1686,10 @@ class SkipUpBlock2D(nn.Module):
|
|
1649 |
# pop res hidden states
|
1650 |
res_hidden_states = res_hidden_states_tuple[-1]
|
1651 |
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
1652 |
-
hidden_states = torch.cat(
|
|
|
1653 |
|
1654 |
-
hidden_states = resnet(hidden_states, temb)
|
1655 |
|
1656 |
if skip_sample is not None:
|
1657 |
skip_sample = self.upsampler(skip_sample)
|
@@ -1668,3 +1706,150 @@ class SkipUpBlock2D(nn.Module):
|
|
1668 |
hidden_states = self.resnet_up(hidden_states, temb)
|
1669 |
|
1670 |
return hidden_states, skip_sample
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
from torch import nn
|
17 |
|
18 |
from .attention import AttentionBlock, DualTransformer2DModel, Transformer2DModel
|
19 |
+
from diffusers.models.resnet import Downsample2D, FirDownsample2D, FirUpsample2D, Upsample2D
|
20 |
|
21 |
|
22 |
def get_down_block(
|
|
|
36 |
use_linear_projection=False,
|
37 |
only_cross_attention=False,
|
38 |
):
|
39 |
+
down_block_type = down_block_type[7:] if down_block_type.startswith(
|
40 |
+
"UNetRes") else down_block_type
|
41 |
if down_block_type == "DownBlock2D":
|
42 |
return DownBlock2D(
|
43 |
num_layers=num_layers,
|
|
|
65 |
)
|
66 |
elif down_block_type == "CrossAttnDownBlock2D":
|
67 |
if cross_attention_dim is None:
|
68 |
+
raise ValueError(
|
69 |
+
"cross_attention_dim must be specified for CrossAttnDownBlock2D")
|
70 |
return CrossAttnDownBlock2D(
|
71 |
num_layers=num_layers,
|
72 |
in_channels=in_channels,
|
|
|
149 |
use_linear_projection=False,
|
150 |
only_cross_attention=False,
|
151 |
):
|
152 |
+
up_block_type = up_block_type[7:] if up_block_type.startswith(
|
153 |
+
"UNetRes") else up_block_type
|
154 |
if up_block_type == "UpBlock2D":
|
155 |
return UpBlock2D(
|
156 |
num_layers=num_layers,
|
|
|
165 |
)
|
166 |
elif up_block_type == "CrossAttnUpBlock2D":
|
167 |
if cross_attention_dim is None:
|
168 |
+
raise ValueError(
|
169 |
+
"cross_attention_dim must be specified for CrossAttnUpBlock2D")
|
170 |
return CrossAttnUpBlock2D(
|
171 |
num_layers=num_layers,
|
172 |
in_channels=in_channels,
|
|
|
262 |
super().__init__()
|
263 |
|
264 |
self.attention_type = attention_type
|
265 |
+
resnet_groups = resnet_groups if resnet_groups is not None else min(
|
266 |
+
in_channels // 4, 32)
|
267 |
|
268 |
# there is always at least one resnet
|
269 |
resnets = [
|
|
|
317 |
hidden_states = attn(hidden_states)
|
318 |
else:
|
319 |
hidden_states = attn(hidden_states, encoder_states)
|
320 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
321 |
|
322 |
return hidden_states
|
323 |
|
|
|
345 |
|
346 |
self.attention_type = attention_type
|
347 |
self.attn_num_head_channels = attn_num_head_channels
|
348 |
+
resnet_groups = resnet_groups if resnet_groups is not None else min(
|
349 |
+
in_channels // 4, 32)
|
350 |
|
351 |
# there is always at least one resnet
|
352 |
resnets = [
|
|
|
426 |
|
427 |
def set_use_memory_efficient_attention_xformers(self, use_memory_efficient_attention_xformers: bool):
|
428 |
for attn in self.attentions:
|
429 |
+
attn._set_use_memory_efficient_attention_xformers(
|
430 |
+
use_memory_efficient_attention_xformers)
|
431 |
|
432 |
def forward(self, hidden_states, temb=None, encoder_hidden_states=None,
|
433 |
text_format_dict={}):
|
434 |
+
hidden_states, _ = self.resnets[0](hidden_states, temb)
|
435 |
for attn, resnet in zip(self.attentions, self.resnets[1:]):
|
436 |
+
hidden_states = attn(hidden_states, encoder_hidden_states,
|
437 |
text_format_dict).sample
|
438 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
439 |
|
440 |
return hidden_states
|
441 |
|
|
|
509 |
output_states = ()
|
510 |
|
511 |
for resnet, attn in zip(self.resnets, self.attentions):
|
512 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
513 |
hidden_states = attn(hidden_states)
|
514 |
output_states += (hidden_states,)
|
515 |
|
|
|
627 |
|
628 |
def set_use_memory_efficient_attention_xformers(self, use_memory_efficient_attention_xformers: bool):
|
629 |
for attn in self.attentions:
|
630 |
+
attn._set_use_memory_efficient_attention_xformers(
|
631 |
+
use_memory_efficient_attention_xformers)
|
632 |
|
633 |
def forward(self, hidden_states, temb=None, encoder_hidden_states=None,
|
634 |
text_format_dict={}):
|
|
|
646 |
|
647 |
return custom_forward
|
648 |
|
|
|
649 |
hidden_states = torch.utils.checkpoint.checkpoint(
|
650 |
+
create_custom_forward(resnet), hidden_states, temb)
|
651 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
652 |
+
create_custom_forward(
|
653 |
+
attn, return_dict=False), hidden_states, encoder_hidden_states,
|
654 |
+
text_format_dict
|
655 |
)[0]
|
656 |
else:
|
657 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
658 |
hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states,
|
659 |
text_format_dict=text_format_dict).sample
|
660 |
|
|
|
733 |
|
734 |
return custom_forward
|
735 |
|
736 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
737 |
+
create_custom_forward(resnet), hidden_states, temb)
|
738 |
else:
|
739 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
740 |
|
741 |
output_states += (hidden_states,)
|
742 |
|
|
|
800 |
|
801 |
def forward(self, hidden_states):
|
802 |
for resnet in self.resnets:
|
803 |
+
hidden_states, _ = resnet(hidden_states, temb=None)
|
804 |
|
805 |
if self.downsamplers is not None:
|
806 |
for downsampler in self.downsamplers:
|
|
|
872 |
|
873 |
def forward(self, hidden_states):
|
874 |
for resnet, attn in zip(self.resnets, self.attentions):
|
875 |
+
hidden_states, _ = resnet(hidden_states, temb=None)
|
876 |
hidden_states = attn(hidden_states)
|
877 |
|
878 |
if self.downsamplers is not None:
|
|
|
948 |
down=True,
|
949 |
kernel="fir",
|
950 |
)
|
951 |
+
self.downsamplers = nn.ModuleList(
|
952 |
+
[FirDownsample2D(out_channels, out_channels=out_channels)])
|
953 |
+
self.skip_conv = nn.Conv2d(
|
954 |
+
3, out_channels, kernel_size=(1, 1), stride=(1, 1))
|
955 |
else:
|
956 |
self.resnet_down = None
|
957 |
self.downsamplers = None
|
|
|
961 |
output_states = ()
|
962 |
|
963 |
for resnet, attn in zip(self.resnets, self.attentions):
|
964 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
965 |
hidden_states = attn(hidden_states)
|
966 |
output_states += (hidden_states,)
|
967 |
|
|
|
1030 |
down=True,
|
1031 |
kernel="fir",
|
1032 |
)
|
1033 |
+
self.downsamplers = nn.ModuleList(
|
1034 |
+
[FirDownsample2D(out_channels, out_channels=out_channels)])
|
1035 |
+
self.skip_conv = nn.Conv2d(
|
1036 |
+
3, out_channels, kernel_size=(1, 1), stride=(1, 1))
|
1037 |
else:
|
1038 |
self.resnet_down = None
|
1039 |
self.downsamplers = None
|
|
|
1043 |
output_states = ()
|
1044 |
|
1045 |
for resnet in self.resnets:
|
1046 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
1047 |
output_states += (hidden_states,)
|
1048 |
|
1049 |
if self.downsamplers is not None:
|
|
|
1084 |
self.attention_type = attention_type
|
1085 |
|
1086 |
for i in range(num_layers):
|
1087 |
+
res_skip_channels = in_channels if (
|
1088 |
+
i == num_layers - 1) else out_channels
|
1089 |
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
1090 |
|
1091 |
resnets.append(
|
|
|
1116 |
self.resnets = nn.ModuleList(resnets)
|
1117 |
|
1118 |
if add_upsample:
|
1119 |
+
self.upsamplers = nn.ModuleList(
|
1120 |
+
[Upsample2D(out_channels, use_conv=True, out_channels=out_channels)])
|
1121 |
else:
|
1122 |
self.upsamplers = None
|
1123 |
|
|
|
1126 |
# pop res hidden states
|
1127 |
res_hidden_states = res_hidden_states_tuple[-1]
|
1128 |
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
1129 |
+
hidden_states = torch.cat(
|
1130 |
+
[hidden_states, res_hidden_states], dim=1)
|
1131 |
|
1132 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
1133 |
hidden_states = attn(hidden_states)
|
1134 |
|
1135 |
if self.upsamplers is not None:
|
|
|
1170 |
self.attn_num_head_channels = attn_num_head_channels
|
1171 |
|
1172 |
for i in range(num_layers):
|
1173 |
+
res_skip_channels = in_channels if (
|
1174 |
+
i == num_layers - 1) else out_channels
|
1175 |
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
1176 |
|
1177 |
resnets.append(
|
|
|
1216 |
self.resnets = nn.ModuleList(resnets)
|
1217 |
|
1218 |
if add_upsample:
|
1219 |
+
self.upsamplers = nn.ModuleList(
|
1220 |
+
[Upsample2D(out_channels, use_conv=True, out_channels=out_channels)])
|
1221 |
else:
|
1222 |
self.upsamplers = None
|
1223 |
|
|
|
1244 |
|
1245 |
def set_use_memory_efficient_attention_xformers(self, use_memory_efficient_attention_xformers: bool):
|
1246 |
for attn in self.attentions:
|
1247 |
+
attn._set_use_memory_efficient_attention_xformers(
|
1248 |
+
use_memory_efficient_attention_xformers)
|
1249 |
|
1250 |
def forward(
|
1251 |
self,
|
|
|
1260 |
# pop res hidden states
|
1261 |
res_hidden_states = res_hidden_states_tuple[-1]
|
1262 |
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
1263 |
+
hidden_states = torch.cat(
|
1264 |
+
[hidden_states, res_hidden_states], dim=1)
|
1265 |
|
1266 |
if self.training and self.gradient_checkpointing:
|
1267 |
|
|
|
1274 |
|
1275 |
return custom_forward
|
1276 |
|
|
|
1277 |
hidden_states = torch.utils.checkpoint.checkpoint(
|
1278 |
+
create_custom_forward(resnet), hidden_states, temb)
|
1279 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
1280 |
+
create_custom_forward(
|
1281 |
+
attn, return_dict=False), hidden_states, encoder_hidden_states,
|
1282 |
+
text_format_dict
|
1283 |
)[0]
|
1284 |
else:
|
1285 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
1286 |
hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states,
|
1287 |
text_format_dict=text_format_dict).sample
|
1288 |
|
|
|
1314 |
resnets = []
|
1315 |
|
1316 |
for i in range(num_layers):
|
1317 |
+
res_skip_channels = in_channels if (
|
1318 |
+
i == num_layers - 1) else out_channels
|
1319 |
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
1320 |
|
1321 |
resnets.append(
|
|
|
1336 |
self.resnets = nn.ModuleList(resnets)
|
1337 |
|
1338 |
if add_upsample:
|
1339 |
+
self.upsamplers = nn.ModuleList(
|
1340 |
+
[Upsample2D(out_channels, use_conv=True, out_channels=out_channels)])
|
1341 |
else:
|
1342 |
self.upsamplers = None
|
1343 |
|
|
|
1348 |
# pop res hidden states
|
1349 |
res_hidden_states = res_hidden_states_tuple[-1]
|
1350 |
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
1351 |
+
hidden_states = torch.cat(
|
1352 |
+
[hidden_states, res_hidden_states], dim=1)
|
1353 |
|
1354 |
if self.training and self.gradient_checkpointing:
|
1355 |
|
|
|
1359 |
|
1360 |
return custom_forward
|
1361 |
|
1362 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
1363 |
+
create_custom_forward(resnet), hidden_states, temb)
|
1364 |
else:
|
1365 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
1366 |
|
1367 |
if self.upsamplers is not None:
|
1368 |
for upsampler in self.upsamplers:
|
|
|
1410 |
self.resnets = nn.ModuleList(resnets)
|
1411 |
|
1412 |
if add_upsample:
|
1413 |
+
self.upsamplers = nn.ModuleList(
|
1414 |
+
[Upsample2D(out_channels, use_conv=True, out_channels=out_channels)])
|
1415 |
else:
|
1416 |
self.upsamplers = None
|
1417 |
|
1418 |
def forward(self, hidden_states):
|
1419 |
for resnet in self.resnets:
|
1420 |
+
hidden_states, _ = resnet(hidden_states, temb=None)
|
1421 |
|
1422 |
if self.upsamplers is not None:
|
1423 |
for upsampler in self.upsamplers:
|
|
|
1477 |
self.resnets = nn.ModuleList(resnets)
|
1478 |
|
1479 |
if add_upsample:
|
1480 |
+
self.upsamplers = nn.ModuleList(
|
1481 |
+
[Upsample2D(out_channels, use_conv=True, out_channels=out_channels)])
|
1482 |
else:
|
1483 |
self.upsamplers = None
|
1484 |
|
1485 |
def forward(self, hidden_states):
|
1486 |
for resnet, attn in zip(self.resnets, self.attentions):
|
1487 |
+
hidden_states, _ = resnet(hidden_states, temb=None)
|
1488 |
hidden_states = attn(hidden_states)
|
1489 |
|
1490 |
if self.upsamplers is not None:
|
|
|
1520 |
self.attention_type = attention_type
|
1521 |
|
1522 |
for i in range(num_layers):
|
1523 |
+
res_skip_channels = in_channels if (
|
1524 |
+
i == num_layers - 1) else out_channels
|
1525 |
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
1526 |
|
1527 |
self.resnets.append(
|
|
|
1530 |
out_channels=out_channels,
|
1531 |
temb_channels=temb_channels,
|
1532 |
eps=resnet_eps,
|
1533 |
+
groups=min(resnet_in_channels +
|
1534 |
+
res_skip_channels // 4, 32),
|
1535 |
groups_out=min(out_channels // 4, 32),
|
1536 |
dropout=dropout,
|
1537 |
time_embedding_norm=resnet_time_scale_shift,
|
|
|
1568 |
up=True,
|
1569 |
kernel="fir",
|
1570 |
)
|
1571 |
+
self.skip_conv = nn.Conv2d(out_channels, 3, kernel_size=(
|
1572 |
+
3, 3), stride=(1, 1), padding=(1, 1))
|
1573 |
self.skip_norm = torch.nn.GroupNorm(
|
1574 |
num_groups=min(out_channels // 4, 32), num_channels=out_channels, eps=resnet_eps, affine=True
|
1575 |
)
|
|
|
1585 |
# pop res hidden states
|
1586 |
res_hidden_states = res_hidden_states_tuple[-1]
|
1587 |
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
1588 |
+
hidden_states = torch.cat(
|
1589 |
+
[hidden_states, res_hidden_states], dim=1)
|
1590 |
|
1591 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
1592 |
|
1593 |
hidden_states = self.attentions[0](hidden_states)
|
1594 |
|
|
|
1630 |
self.resnets = nn.ModuleList([])
|
1631 |
|
1632 |
for i in range(num_layers):
|
1633 |
+
res_skip_channels = in_channels if (
|
1634 |
+
i == num_layers - 1) else out_channels
|
1635 |
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
1636 |
|
1637 |
self.resnets.append(
|
|
|
1640 |
out_channels=out_channels,
|
1641 |
temb_channels=temb_channels,
|
1642 |
eps=resnet_eps,
|
1643 |
+
groups=min(
|
1644 |
+
(resnet_in_channels + res_skip_channels) // 4, 32),
|
1645 |
groups_out=min(out_channels // 4, 32),
|
1646 |
dropout=dropout,
|
1647 |
time_embedding_norm=resnet_time_scale_shift,
|
|
|
1669 |
up=True,
|
1670 |
kernel="fir",
|
1671 |
)
|
1672 |
+
self.skip_conv = nn.Conv2d(out_channels, 3, kernel_size=(
|
1673 |
+
3, 3), stride=(1, 1), padding=(1, 1))
|
1674 |
self.skip_norm = torch.nn.GroupNorm(
|
1675 |
num_groups=min(out_channels // 4, 32), num_channels=out_channels, eps=resnet_eps, affine=True
|
1676 |
)
|
|
|
1686 |
# pop res hidden states
|
1687 |
res_hidden_states = res_hidden_states_tuple[-1]
|
1688 |
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
1689 |
+
hidden_states = torch.cat(
|
1690 |
+
[hidden_states, res_hidden_states], dim=1)
|
1691 |
|
1692 |
+
hidden_states, _ = resnet(hidden_states, temb)
|
1693 |
|
1694 |
if skip_sample is not None:
|
1695 |
skip_sample = self.upsampler(skip_sample)
|
|
|
1706 |
hidden_states = self.resnet_up(hidden_states, temb)
|
1707 |
|
1708 |
return hidden_states, skip_sample
|
1709 |
+
|
1710 |
+
|
1711 |
+
class ResnetBlock2D(nn.Module):
|
1712 |
+
def __init__(
|
1713 |
+
self,
|
1714 |
+
*,
|
1715 |
+
in_channels,
|
1716 |
+
out_channels=None,
|
1717 |
+
conv_shortcut=False,
|
1718 |
+
dropout=0.0,
|
1719 |
+
temb_channels=512,
|
1720 |
+
groups=32,
|
1721 |
+
groups_out=None,
|
1722 |
+
pre_norm=True,
|
1723 |
+
eps=1e-6,
|
1724 |
+
non_linearity="swish",
|
1725 |
+
time_embedding_norm="default",
|
1726 |
+
kernel=None,
|
1727 |
+
output_scale_factor=1.0,
|
1728 |
+
use_in_shortcut=None,
|
1729 |
+
up=False,
|
1730 |
+
down=False,
|
1731 |
+
):
|
1732 |
+
super().__init__()
|
1733 |
+
self.pre_norm = pre_norm
|
1734 |
+
self.pre_norm = True
|
1735 |
+
self.in_channels = in_channels
|
1736 |
+
out_channels = in_channels if out_channels is None else out_channels
|
1737 |
+
self.out_channels = out_channels
|
1738 |
+
self.use_conv_shortcut = conv_shortcut
|
1739 |
+
self.time_embedding_norm = time_embedding_norm
|
1740 |
+
self.up = up
|
1741 |
+
self.down = down
|
1742 |
+
self.output_scale_factor = output_scale_factor
|
1743 |
+
|
1744 |
+
if groups_out is None:
|
1745 |
+
groups_out = groups
|
1746 |
+
|
1747 |
+
self.norm1 = torch.nn.GroupNorm(
|
1748 |
+
num_groups=groups, num_channels=in_channels, eps=eps, affine=True)
|
1749 |
+
|
1750 |
+
self.conv1 = torch.nn.Conv2d(
|
1751 |
+
in_channels, out_channels, kernel_size=3, stride=1, padding=1)
|
1752 |
+
|
1753 |
+
if temb_channels is not None:
|
1754 |
+
if self.time_embedding_norm == "default":
|
1755 |
+
time_emb_proj_out_channels = out_channels
|
1756 |
+
elif self.time_embedding_norm == "scale_shift":
|
1757 |
+
time_emb_proj_out_channels = out_channels * 2
|
1758 |
+
else:
|
1759 |
+
raise ValueError(
|
1760 |
+
f"unknown time_embedding_norm : {self.time_embedding_norm} ")
|
1761 |
+
|
1762 |
+
self.time_emb_proj = torch.nn.Linear(
|
1763 |
+
temb_channels, time_emb_proj_out_channels)
|
1764 |
+
else:
|
1765 |
+
self.time_emb_proj = None
|
1766 |
+
|
1767 |
+
self.norm2 = torch.nn.GroupNorm(
|
1768 |
+
num_groups=groups_out, num_channels=out_channels, eps=eps, affine=True)
|
1769 |
+
self.dropout = torch.nn.Dropout(dropout)
|
1770 |
+
self.conv2 = torch.nn.Conv2d(
|
1771 |
+
out_channels, out_channels, kernel_size=3, stride=1, padding=1)
|
1772 |
+
|
1773 |
+
if non_linearity == "swish":
|
1774 |
+
self.nonlinearity = lambda x: F.silu(x)
|
1775 |
+
elif non_linearity == "mish":
|
1776 |
+
self.nonlinearity = Mish()
|
1777 |
+
elif non_linearity == "silu":
|
1778 |
+
self.nonlinearity = nn.SiLU()
|
1779 |
+
|
1780 |
+
self.upsample = self.downsample = None
|
1781 |
+
if self.up:
|
1782 |
+
if kernel == "fir":
|
1783 |
+
fir_kernel = (1, 3, 3, 1)
|
1784 |
+
self.upsample = lambda x: upsample_2d(x, kernel=fir_kernel)
|
1785 |
+
elif kernel == "sde_vp":
|
1786 |
+
self.upsample = partial(
|
1787 |
+
F.interpolate, scale_factor=2.0, mode="nearest")
|
1788 |
+
else:
|
1789 |
+
self.upsample = Upsample2D(in_channels, use_conv=False)
|
1790 |
+
elif self.down:
|
1791 |
+
if kernel == "fir":
|
1792 |
+
fir_kernel = (1, 3, 3, 1)
|
1793 |
+
self.downsample = lambda x: downsample_2d(x, kernel=fir_kernel)
|
1794 |
+
elif kernel == "sde_vp":
|
1795 |
+
self.downsample = partial(
|
1796 |
+
F.avg_pool2d, kernel_size=2, stride=2)
|
1797 |
+
else:
|
1798 |
+
self.downsample = Downsample2D(
|
1799 |
+
in_channels, use_conv=False, padding=1, name="op")
|
1800 |
+
|
1801 |
+
self.use_in_shortcut = self.in_channels != self.out_channels if use_in_shortcut is None else use_in_shortcut
|
1802 |
+
|
1803 |
+
self.conv_shortcut = None
|
1804 |
+
if self.use_in_shortcut:
|
1805 |
+
self.conv_shortcut = torch.nn.Conv2d(
|
1806 |
+
in_channels, out_channels, kernel_size=1, stride=1, padding=0)
|
1807 |
+
|
1808 |
+
def forward(self, input_tensor, temb, inject_states=None):
|
1809 |
+
hidden_states = input_tensor
|
1810 |
+
|
1811 |
+
hidden_states = self.norm1(hidden_states)
|
1812 |
+
hidden_states = self.nonlinearity(hidden_states)
|
1813 |
+
|
1814 |
+
if self.upsample is not None:
|
1815 |
+
# upsample_nearest_nhwc fails with large batch sizes. see https://github.com/huggingface/diffusers/issues/984
|
1816 |
+
if hidden_states.shape[0] >= 64:
|
1817 |
+
input_tensor = input_tensor.contiguous()
|
1818 |
+
hidden_states = hidden_states.contiguous()
|
1819 |
+
input_tensor = self.upsample(input_tensor)
|
1820 |
+
hidden_states = self.upsample(hidden_states)
|
1821 |
+
elif self.downsample is not None:
|
1822 |
+
input_tensor = self.downsample(input_tensor)
|
1823 |
+
hidden_states = self.downsample(hidden_states)
|
1824 |
+
|
1825 |
+
hidden_states = self.conv1(hidden_states)
|
1826 |
+
|
1827 |
+
if temb is not None:
|
1828 |
+
temb = self.time_emb_proj(self.nonlinearity(temb))[
|
1829 |
+
:, :, None, None]
|
1830 |
+
|
1831 |
+
if temb is not None and self.time_embedding_norm == "default":
|
1832 |
+
hidden_states = hidden_states + temb
|
1833 |
+
|
1834 |
+
hidden_states = self.norm2(hidden_states)
|
1835 |
+
|
1836 |
+
if temb is not None and self.time_embedding_norm == "scale_shift":
|
1837 |
+
scale, shift = torch.chunk(temb, 2, dim=1)
|
1838 |
+
hidden_states = hidden_states * (1 + scale) + shift
|
1839 |
+
|
1840 |
+
hidden_states = self.nonlinearity(hidden_states)
|
1841 |
+
|
1842 |
+
hidden_states = self.dropout(hidden_states)
|
1843 |
+
hidden_states = self.conv2(hidden_states)
|
1844 |
+
|
1845 |
+
if self.conv_shortcut is not None:
|
1846 |
+
input_tensor = self.conv_shortcut(input_tensor)
|
1847 |
+
|
1848 |
+
if inject_states is not None:
|
1849 |
+
output_tensor = (input_tensor + inject_states) / \
|
1850 |
+
self.output_scale_factor
|
1851 |
+
else:
|
1852 |
+
output_tensor = (input_tensor + hidden_states) / \
|
1853 |
+
self.output_scale_factor
|
1854 |
+
|
1855 |
+
return output_tensor, hidden_states
|
utils/attention_utils.py
CHANGED
@@ -6,7 +6,46 @@ import seaborn as sns
|
|
6 |
import torch
|
7 |
import torchvision
|
8 |
|
9 |
-
from
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
|
11 |
|
12 |
def split_attention_maps_over_steps(attention_maps):
|
@@ -37,7 +76,7 @@ def split_attention_maps_over_steps(attention_maps):
|
|
37 |
|
38 |
def plot_attention_maps(atten_map_list, obj_tokens, save_dir, seed, tokens_vis=None):
|
39 |
atten_names = ['presoftmax', 'postsoftmax', 'postsoftmax_erosion']
|
40 |
-
for i,
|
41 |
n_obj = len(attn_map)
|
42 |
plt.figure()
|
43 |
plt.clf()
|
@@ -63,6 +102,7 @@ def plot_attention_maps(atten_map_list, obj_tokens, save_dir, seed, tokens_vis=N
|
|
63 |
cmap=cmap, vmin=vmin, vmax=vmax
|
64 |
)
|
65 |
axs[tid].set_axis_off()
|
|
|
66 |
if tokens_vis is not None:
|
67 |
if tid == n_obj-1:
|
68 |
axs_xlabel = 'other tokens'
|
@@ -79,13 +119,14 @@ def plot_attention_maps(atten_map_list, obj_tokens, save_dir, seed, tokens_vis=N
|
|
79 |
canvas = fig.canvas
|
80 |
canvas.draw()
|
81 |
width, height = canvas.get_width_height()
|
82 |
-
img = np.frombuffer(canvas.tostring_rgb(),
|
|
|
83 |
|
84 |
fig.tight_layout()
|
85 |
return img
|
86 |
|
87 |
|
88 |
-
def
|
89 |
r"""Function to visualize attention maps.
|
90 |
Args:
|
91 |
save_dir (str): Path to save attention maps
|
@@ -98,25 +139,6 @@ def get_token_maps(attention_maps, save_dir, width, height, obj_tokens, seed=0,
|
|
98 |
attention_maps
|
99 |
)
|
100 |
|
101 |
-
selected_layers = [
|
102 |
-
# 'down_blocks.0.attentions.0.transformer_blocks.0.attn2',
|
103 |
-
# 'down_blocks.0.attentions.1.transformer_blocks.0.attn2',
|
104 |
-
'down_blocks.1.attentions.0.transformer_blocks.0.attn2',
|
105 |
-
# 'down_blocks.1.attentions.1.transformer_blocks.0.attn2',
|
106 |
-
'down_blocks.2.attentions.0.transformer_blocks.0.attn2',
|
107 |
-
'down_blocks.2.attentions.1.transformer_blocks.0.attn2',
|
108 |
-
'mid_block.attentions.0.transformer_blocks.0.attn2',
|
109 |
-
'up_blocks.1.attentions.0.transformer_blocks.0.attn2',
|
110 |
-
'up_blocks.1.attentions.1.transformer_blocks.0.attn2',
|
111 |
-
'up_blocks.1.attentions.2.transformer_blocks.0.attn2',
|
112 |
-
# 'up_blocks.2.attentions.0.transformer_blocks.0.attn2',
|
113 |
-
'up_blocks.2.attentions.1.transformer_blocks.0.attn2',
|
114 |
-
# 'up_blocks.2.attentions.2.transformer_blocks.0.attn2',
|
115 |
-
# 'up_blocks.3.attentions.0.transformer_blocks.0.attn2',
|
116 |
-
# 'up_blocks.3.attentions.1.transformer_blocks.0.attn2',
|
117 |
-
# 'up_blocks.3.attentions.2.transformer_blocks.0.attn2'
|
118 |
-
]
|
119 |
-
|
120 |
nsteps = len(attention_maps_cond)
|
121 |
hw_ori = width * height
|
122 |
|
@@ -128,7 +150,7 @@ def get_token_maps(attention_maps, save_dir, width, height, obj_tokens, seed=0,
|
|
128 |
attention_maps_cur = attention_maps_cond[step_num]
|
129 |
|
130 |
for layer in attention_maps_cur.keys():
|
131 |
-
if step_num < 10 or layer not in
|
132 |
continue
|
133 |
|
134 |
attention_ind = attention_maps_cur[layer].cpu()
|
@@ -179,7 +201,107 @@ def get_token_maps(attention_maps, save_dir, width, height, obj_tokens, seed=0,
|
|
179 |
attention_maps_averaged_normalized[i:i+1] for i in range(attention_maps_averaged_normalized.shape[0])]
|
180 |
|
181 |
token_maps_vis = plot_attention_maps([attention_maps_averaged, attention_maps_averaged_normalized],
|
182 |
-
|
183 |
attention_maps_averaged_normalized = [attn_mask.unsqueeze(1).repeat(
|
184 |
[1, 4, 1, 1]).cuda() for attn_mask in attention_maps_averaged_normalized]
|
185 |
return attention_maps_averaged_normalized, token_maps_vis
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
import torch
|
7 |
import torchvision
|
8 |
|
9 |
+
from sklearn.cluster import KMeans
|
10 |
+
|
11 |
+
SelfAttentionLayers = [
|
12 |
+
# 'down_blocks.0.attentions.0.transformer_blocks.0.attn1',
|
13 |
+
# 'down_blocks.0.attentions.1.transformer_blocks.0.attn1',
|
14 |
+
'down_blocks.1.attentions.0.transformer_blocks.0.attn1',
|
15 |
+
# 'down_blocks.1.attentions.1.transformer_blocks.0.attn1',
|
16 |
+
'down_blocks.2.attentions.0.transformer_blocks.0.attn1',
|
17 |
+
'down_blocks.2.attentions.1.transformer_blocks.0.attn1',
|
18 |
+
'mid_block.attentions.0.transformer_blocks.0.attn1',
|
19 |
+
'up_blocks.1.attentions.0.transformer_blocks.0.attn1',
|
20 |
+
'up_blocks.1.attentions.1.transformer_blocks.0.attn1',
|
21 |
+
'up_blocks.1.attentions.2.transformer_blocks.0.attn1',
|
22 |
+
# 'up_blocks.2.attentions.0.transformer_blocks.0.attn1',
|
23 |
+
'up_blocks.2.attentions.1.transformer_blocks.0.attn1',
|
24 |
+
# 'up_blocks.2.attentions.2.transformer_blocks.0.attn1',
|
25 |
+
# 'up_blocks.3.attentions.0.transformer_blocks.0.attn1',
|
26 |
+
# 'up_blocks.3.attentions.1.transformer_blocks.0.attn1',
|
27 |
+
# 'up_blocks.3.attentions.2.transformer_blocks.0.attn1',
|
28 |
+
]
|
29 |
+
|
30 |
+
|
31 |
+
CrossAttentionLayers = [
|
32 |
+
# 'down_blocks.0.attentions.0.transformer_blocks.0.attn2',
|
33 |
+
# 'down_blocks.0.attentions.1.transformer_blocks.0.attn2',
|
34 |
+
'down_blocks.1.attentions.0.transformer_blocks.0.attn2',
|
35 |
+
# 'down_blocks.1.attentions.1.transformer_blocks.0.attn2',
|
36 |
+
'down_blocks.2.attentions.0.transformer_blocks.0.attn2',
|
37 |
+
'down_blocks.2.attentions.1.transformer_blocks.0.attn2',
|
38 |
+
'mid_block.attentions.0.transformer_blocks.0.attn2',
|
39 |
+
'up_blocks.1.attentions.0.transformer_blocks.0.attn2',
|
40 |
+
'up_blocks.1.attentions.1.transformer_blocks.0.attn2',
|
41 |
+
'up_blocks.1.attentions.2.transformer_blocks.0.attn2',
|
42 |
+
# 'up_blocks.2.attentions.0.transformer_blocks.0.attn2',
|
43 |
+
'up_blocks.2.attentions.1.transformer_blocks.0.attn2',
|
44 |
+
# 'up_blocks.2.attentions.2.transformer_blocks.0.attn2',
|
45 |
+
# 'up_blocks.3.attentions.0.transformer_blocks.0.attn2',
|
46 |
+
# 'up_blocks.3.attentions.1.transformer_blocks.0.attn2',
|
47 |
+
# 'up_blocks.3.attentions.2.transformer_blocks.0.attn2'
|
48 |
+
]
|
49 |
|
50 |
|
51 |
def split_attention_maps_over_steps(attention_maps):
|
|
|
76 |
|
77 |
def plot_attention_maps(atten_map_list, obj_tokens, save_dir, seed, tokens_vis=None):
|
78 |
atten_names = ['presoftmax', 'postsoftmax', 'postsoftmax_erosion']
|
79 |
+
for i, attn_map in enumerate(atten_map_list):
|
80 |
n_obj = len(attn_map)
|
81 |
plt.figure()
|
82 |
plt.clf()
|
|
|
102 |
cmap=cmap, vmin=vmin, vmax=vmax
|
103 |
)
|
104 |
axs[tid].set_axis_off()
|
105 |
+
|
106 |
if tokens_vis is not None:
|
107 |
if tid == n_obj-1:
|
108 |
axs_xlabel = 'other tokens'
|
|
|
119 |
canvas = fig.canvas
|
120 |
canvas.draw()
|
121 |
width, height = canvas.get_width_height()
|
122 |
+
img = np.frombuffer(canvas.tostring_rgb(),
|
123 |
+
dtype='uint8').reshape((height, width, 3))
|
124 |
|
125 |
fig.tight_layout()
|
126 |
return img
|
127 |
|
128 |
|
129 |
+
def get_token_maps_deprecated(attention_maps, save_dir, width, height, obj_tokens, seed=0, tokens_vis=None):
|
130 |
r"""Function to visualize attention maps.
|
131 |
Args:
|
132 |
save_dir (str): Path to save attention maps
|
|
|
139 |
attention_maps
|
140 |
)
|
141 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
142 |
nsteps = len(attention_maps_cond)
|
143 |
hw_ori = width * height
|
144 |
|
|
|
150 |
attention_maps_cur = attention_maps_cond[step_num]
|
151 |
|
152 |
for layer in attention_maps_cur.keys():
|
153 |
+
if step_num < 10 or layer not in CrossAttentionLayers:
|
154 |
continue
|
155 |
|
156 |
attention_ind = attention_maps_cur[layer].cpu()
|
|
|
201 |
attention_maps_averaged_normalized[i:i+1] for i in range(attention_maps_averaged_normalized.shape[0])]
|
202 |
|
203 |
token_maps_vis = plot_attention_maps([attention_maps_averaged, attention_maps_averaged_normalized],
|
204 |
+
obj_tokens, save_dir, seed, tokens_vis)
|
205 |
attention_maps_averaged_normalized = [attn_mask.unsqueeze(1).repeat(
|
206 |
[1, 4, 1, 1]).cuda() for attn_mask in attention_maps_averaged_normalized]
|
207 |
return attention_maps_averaged_normalized, token_maps_vis
|
208 |
+
|
209 |
+
|
210 |
+
def get_token_maps(selfattn_maps, crossattn_maps, n_maps, save_dir, width, height, obj_tokens, kmeans_seed=0, tokens_vis=None,
|
211 |
+
preprocess=False, segment_threshold=0.30, num_segments=9, return_vis=False):
|
212 |
+
r"""Function to visualize attention maps.
|
213 |
+
Args:
|
214 |
+
save_dir (str): Path to save attention maps
|
215 |
+
batch_size (int): Batch size
|
216 |
+
sampler_order (int): Sampler order
|
217 |
+
"""
|
218 |
+
|
219 |
+
# create the segmentation mask using self-attention maps
|
220 |
+
resolution = 32
|
221 |
+
attn_maps_1024 = {8: [], 16: [], 32: []}
|
222 |
+
for attn_map in selfattn_maps.values():
|
223 |
+
resolution_map = np.sqrt(attn_map.shape[1]).astype(int)
|
224 |
+
attn_map = attn_map.reshape(
|
225 |
+
1, resolution_map, resolution_map, resolution_map**2).permute([3, 0, 1, 2])
|
226 |
+
attn_map = torch.nn.functional.interpolate(attn_map, (resolution, resolution),
|
227 |
+
mode='bicubic', antialias=True)
|
228 |
+
attn_maps_1024[resolution_map].append(attn_map.permute([1, 2, 3, 0]).reshape(
|
229 |
+
1, resolution**2, resolution_map**2))
|
230 |
+
attn_maps_1024 = torch.cat([torch.cat(v).mean(0).cpu()
|
231 |
+
for v in attn_maps_1024.values()], -1).numpy()
|
232 |
+
kmeans = KMeans(n_clusters=num_segments,
|
233 |
+
n_init=10).fit(attn_maps_1024)
|
234 |
+
clusters = kmeans.labels_
|
235 |
+
clusters = clusters.reshape(resolution, resolution)
|
236 |
+
fig = plt.figure()
|
237 |
+
plt.imshow(clusters)
|
238 |
+
plt.axis('off')
|
239 |
+
plt.savefig(os.path.join(save_dir, 'segmentation_k%d.jpg' % (num_segments)),
|
240 |
+
bbox_inches='tight', pad_inches=0)
|
241 |
+
if return_vis:
|
242 |
+
canvas = fig.canvas
|
243 |
+
canvas.draw()
|
244 |
+
cav_width, cav_height = canvas.get_width_height()
|
245 |
+
segments_vis = np.frombuffer(canvas.tostring_rgb(),
|
246 |
+
dtype='uint8').reshape((cav_height, cav_width, 3))
|
247 |
+
|
248 |
+
plt.close()
|
249 |
+
|
250 |
+
# label the segmentation mask using cross-attention maps
|
251 |
+
cross_attn_maps_1024 = []
|
252 |
+
for attn_map in crossattn_maps.values():
|
253 |
+
resolution_map = np.sqrt(attn_map.shape[1]).astype(int)
|
254 |
+
attn_map = attn_map.reshape(
|
255 |
+
1, resolution_map, resolution_map, -1).permute([0, 3, 1, 2])
|
256 |
+
attn_map = torch.nn.functional.interpolate(attn_map, (resolution, resolution),
|
257 |
+
mode='bicubic', antialias=True)
|
258 |
+
cross_attn_maps_1024.append(attn_map.permute([0, 2, 3, 1]))
|
259 |
+
|
260 |
+
cross_attn_maps_1024 = torch.cat(
|
261 |
+
cross_attn_maps_1024).mean(0).cpu().numpy()
|
262 |
+
normalized_span_maps = []
|
263 |
+
for token_ids in obj_tokens:
|
264 |
+
span_token_maps = cross_attn_maps_1024[:, :, token_ids.numpy()]
|
265 |
+
normalized_span_map = np.zeros_like(span_token_maps)
|
266 |
+
for i in range(span_token_maps.shape[-1]):
|
267 |
+
curr_noun_map = span_token_maps[:, :, i]
|
268 |
+
normalized_span_map[:, :, i] = (
|
269 |
+
curr_noun_map - np.abs(curr_noun_map.min())) / curr_noun_map.max()
|
270 |
+
normalized_span_maps.append(normalized_span_map)
|
271 |
+
foreground_token_maps = [np.zeros([clusters.shape[0], clusters.shape[1]]).squeeze(
|
272 |
+
) for normalized_span_map in normalized_span_maps]
|
273 |
+
background_map = np.zeros([clusters.shape[0], clusters.shape[1]]).squeeze()
|
274 |
+
for c in range(num_segments):
|
275 |
+
cluster_mask = np.zeros_like(clusters)
|
276 |
+
cluster_mask[clusters == c] = 1.
|
277 |
+
is_foreground = False
|
278 |
+
for normalized_span_map, foreground_nouns_map, token_ids in zip(normalized_span_maps, foreground_token_maps, obj_tokens):
|
279 |
+
score_maps = [cluster_mask * normalized_span_map[:, :, i]
|
280 |
+
for i in range(len(token_ids))]
|
281 |
+
scores = [score_map.sum() / cluster_mask.sum()
|
282 |
+
for score_map in score_maps]
|
283 |
+
if max(scores) > segment_threshold:
|
284 |
+
foreground_nouns_map += cluster_mask
|
285 |
+
is_foreground = True
|
286 |
+
if not is_foreground:
|
287 |
+
background_map += cluster_mask
|
288 |
+
foreground_token_maps.append(background_map)
|
289 |
+
|
290 |
+
# resize the token maps and visualization
|
291 |
+
resized_token_maps = torch.cat([torch.nn.functional.interpolate(torch.from_numpy(token_map).unsqueeze(0).unsqueeze(
|
292 |
+
0), (height, width), mode='bicubic', antialias=True)[0] for token_map in foreground_token_maps]).clamp(0, 1)
|
293 |
+
|
294 |
+
resized_token_maps = resized_token_maps / \
|
295 |
+
(resized_token_maps.sum(0, True)+1e-8)
|
296 |
+
resized_token_maps = [token_map.unsqueeze(
|
297 |
+
0) for token_map in resized_token_maps]
|
298 |
+
foreground_token_maps = [token_map[None, :, :]
|
299 |
+
for token_map in foreground_token_maps]
|
300 |
+
token_maps_vis = plot_attention_maps([foreground_token_maps, resized_token_maps], obj_tokens,
|
301 |
+
save_dir, kmeans_seed, tokens_vis)
|
302 |
+
resized_token_maps = [token_map.unsqueeze(1).repeat(
|
303 |
+
[1, 4, 1, 1]).to(attn_map.dtype).cuda() for token_map in resized_token_maps]
|
304 |
+
if return_vis:
|
305 |
+
return resized_token_maps, segments_vis, token_maps_vis
|
306 |
+
else:
|
307 |
+
return resized_token_maps
|
utils/richtext_utils.py
CHANGED
@@ -27,7 +27,7 @@ def seed_everything(seed):
|
|
27 |
torch.cuda.manual_seed(seed)
|
28 |
|
29 |
|
30 |
-
def hex_to_rgb(hex_string, return_nearest_color=False
|
31 |
r"""
|
32 |
Covert Hex triplet to RGB triplet.
|
33 |
"""
|
@@ -40,8 +40,8 @@ def hex_to_rgb(hex_string, return_nearest_color=False, device='cuda'):
|
|
40 |
rgb = torch.FloatTensor((red, green, blue))[None, :, None, None]/255.
|
41 |
if return_nearest_color:
|
42 |
nearest_color = find_nearest_color(rgb)
|
43 |
-
return rgb.
|
44 |
-
return rgb.
|
45 |
|
46 |
|
47 |
def find_nearest_color(rgb):
|
@@ -56,7 +56,7 @@ def find_nearest_color(rgb):
|
|
56 |
return nearest_color
|
57 |
|
58 |
|
59 |
-
def font2style(font
|
60 |
r"""
|
61 |
Convert the font name to the style name.
|
62 |
"""
|
@@ -71,7 +71,7 @@ def font2style(font, device='cuda'):
|
|
71 |
'Akronim': 'Abstract Cubism, Pablo Picasso', }[font]
|
72 |
|
73 |
|
74 |
-
def parse_json(json_str
|
75 |
r"""
|
76 |
Convert the JSON string to attributes.
|
77 |
"""
|
@@ -121,7 +121,7 @@ def parse_json(json_str, device):
|
|
121 |
if 'color' in span['attributes']:
|
122 |
use_grad_guidance = True
|
123 |
color_rgb, nearest_color = hex_to_rgb(
|
124 |
-
span['attributes']['color'], True
|
125 |
if prev_color_rgb == color_rgb:
|
126 |
prev_text_prompt = color_text_prompts[-1]
|
127 |
color_text_prompts[-1] = prev_text_prompt + \
|
@@ -197,8 +197,8 @@ def get_attention_control_input(model, base_tokens, size_text_prompts_and_sizes)
|
|
197 |
word_pos.append(base_tokens.index(size_token)+1)
|
198 |
font_sizes.append(font_size)
|
199 |
if len(word_pos) > 0:
|
200 |
-
word_pos = torch.LongTensor(word_pos).
|
201 |
-
font_sizes = torch.FloatTensor(font_sizes).
|
202 |
else:
|
203 |
word_pos = None
|
204 |
font_sizes = None
|
|
|
27 |
torch.cuda.manual_seed(seed)
|
28 |
|
29 |
|
30 |
+
def hex_to_rgb(hex_string, return_nearest_color=False):
|
31 |
r"""
|
32 |
Covert Hex triplet to RGB triplet.
|
33 |
"""
|
|
|
40 |
rgb = torch.FloatTensor((red, green, blue))[None, :, None, None]/255.
|
41 |
if return_nearest_color:
|
42 |
nearest_color = find_nearest_color(rgb)
|
43 |
+
return rgb.cuda(), nearest_color
|
44 |
+
return rgb.cuda()
|
45 |
|
46 |
|
47 |
def find_nearest_color(rgb):
|
|
|
56 |
return nearest_color
|
57 |
|
58 |
|
59 |
+
def font2style(font):
|
60 |
r"""
|
61 |
Convert the font name to the style name.
|
62 |
"""
|
|
|
71 |
'Akronim': 'Abstract Cubism, Pablo Picasso', }[font]
|
72 |
|
73 |
|
74 |
+
def parse_json(json_str):
|
75 |
r"""
|
76 |
Convert the JSON string to attributes.
|
77 |
"""
|
|
|
121 |
if 'color' in span['attributes']:
|
122 |
use_grad_guidance = True
|
123 |
color_rgb, nearest_color = hex_to_rgb(
|
124 |
+
span['attributes']['color'], True)
|
125 |
if prev_color_rgb == color_rgb:
|
126 |
prev_text_prompt = color_text_prompts[-1]
|
127 |
color_text_prompts[-1] = prev_text_prompt + \
|
|
|
197 |
word_pos.append(base_tokens.index(size_token)+1)
|
198 |
font_sizes.append(font_size)
|
199 |
if len(word_pos) > 0:
|
200 |
+
word_pos = torch.LongTensor(word_pos).cuda()
|
201 |
+
font_sizes = torch.FloatTensor(font_sizes).cuda()
|
202 |
else:
|
203 |
word_pos = None
|
204 |
font_sizes = None
|