Spaces:
Running
on
Zero
Running
on
Zero
File size: 10,693 Bytes
1de8821 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 |
import os
import cv2
import torch
import spaces
import imageio
import numpy as np
import gradio as gr
torch.jit.script = lambda f: f
import argparse
from utils.batch_inference import (
BSRInferenceLoop, BIDInferenceLoop
)
# import subprocess
# subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
device = 'cuda' if torch.cuda.is_available() else 'cpu'
def get_example(task):
case = {
"dn": [
['examples/bus.mp4',],
['examples/koala.mp4',],
['examples/flamingo.mp4',],
['examples/rhino.mp4',],
['examples/elephant.mp4',],
['examples/sheep.mp4',],
['examples/dog-agility.mp4',],
# ['examples/dog-gooses.mp4',],
],
"sr": [
['examples/bus_sr.mp4',],
['examples/koala_sr.mp4',],
['examples/flamingo_sr.mp4',],
['examples/rhino_sr.mp4',],
['examples/elephant_sr.mp4',],
['examples/sheep_sr.mp4',],
['examples/dog-agility_sr.mp4',],
# ['examples/dog-gooses_sr.mp4',],
]
}
return case[task]
def update_prompt(input_video):
video_name = input_video.split('/')[-1]
return set_default_prompt(video_name)
# Map videos to corresponding images
video_to_image = {
'bus.mp4': ['examples_frames/bus'],
'koala.mp4': ['examples_frames/koala'],
'dog-gooses.mp4': ['examples_frames/dog-gooses'],
'flamingo.mp4': ['examples_frames/flamingo'],
'rhino.mp4': ['examples_frames/rhino'],
'elephant.mp4': ['examples_frames/elephant'],
'sheep.mp4': ['examples_frames/sheep'],
'dog-agility.mp4': ['examples_frames/dog-agility'],
'bus_sr.mp4': ['examples_frames/bus_sr'],
'koala_sr.mp4': ['examples_frames/koala_sr'],
'dog-gooses_sr.mp4': ['examples_frames/dog_gooses_sr'],
'flamingo_sr.mp4': ['examples_frames/flamingo_sr'],
'rhino_sr.mp4': ['examples_frames/rhino_sr'],
'elephant_sr.mp4': ['examples_frames/elephant_sr'],
'sheep_sr.mp4': ['examples_frames/sheep_sr'],
'dog-agility_sr.mp4': ['examples_frames/dog-agility_sr'],
}
def images_to_video(image_list, output_path, fps=10):
# Convert PIL Images to numpy arrays
frames = [np.array(img).astype(np.uint8) for img in image_list]
frames = frames[:20]
# Create video writer
writer = imageio.get_writer(output_path, fps=fps, codec='libx264')
for frame in frames:
writer.append_data(frame)
writer.close()
@spaces.GPU(duration=120)
def DiffBIR_restore(input_video, prompt, sr_ratio, n_frames, n_steps, guidance_scale, seed, n_prompt, task):
video_name = input_video.split('/')[-1]
if video_name in video_to_image:
frames_path = video_to_image[video_name][0]
else:
return None
print(f"[INFO] input_video: {input_video}")
print(f"[INFO] Frames path: {frames_path}")
args = argparse.Namespace()
# args.task = True, choices=["sr", "dn", "fr", "fr_bg"]
args.task = task
args.upscale = sr_ratio
### sampling parameters
args.steps = n_steps
args.better_start = True
args.tiled = False
args.tile_size = 512
args.tile_stride = 256
args.pos_prompt = prompt
args.neg_prompt = n_prompt
args.cfg_scale = guidance_scale
### input parameters
args.input = frames_path
args.n_samples = 1
args.batch_size = 10
args.final_size = (480, 854)
args.config = "configs/inference/my_cldm.yaml"
### guidance parameters
args.guidance = False
args.g_loss = "w_mse"
args.g_scale = 0.0
args.g_start = 1001
args.g_stop = -1
args.g_space = "latent"
args.g_repeat = 1
### output parameters
args.output = " "
### common parameters
args.seed = seed
args.device = "cuda"
args.n_frames = n_frames
### latent control parameters
args.warp_period = [0, 0.1]
args.merge_period = [0, 0]
args.ToMe_period = [0, 1]
args.merge_ratio = [0.6, 0]
if args.task == "sr":
restored_vid_path = BSRInferenceLoop(args).run()
elif args.task == "dn":
restored_vid_path = BIDInferenceLoop(args).run()
torch.cuda.empty_cache()
return restored_vid_path
########
# demo #
########
intro = """
<div style="text-align:center">
<h1 style="font-weight: 1400; text-align: center; margin-bottom: 7px;">
DiffIR2VR - <small>Zero-Shot Video Restoration</small>
</h1>
<span>[<a target="_blank" href="https://jimmycv07.github.io/DiffIR2VR_web/">Project page</a>] [<a target="_blank" href="https://huggingface.co/papers/2406.06523">arXiv</a>]</span>
<div style="display:flex; justify-content: center;margin-top: 0.5em">Note that this page is a limited demo of DiffIR2VR. For more configurations, please visit our GitHub page. The code will be released soon!</div>
</div>
"""
with gr.Blocks(css="style.css") as demo:
gr.HTML(intro)
with gr.Tab(label="Super-resolution with DiffBIR"):
with gr.Row():
input_video = gr.Video(label="Input Video")
output_video = gr.Video(label="Restored Video", interactive=False)
with gr.Row():
run_button = gr.Button("Restore your video !", visible=True)
with gr.Accordion('Advanced options', open=False):
prompt = gr.Textbox(
label="Prompt",
max_lines=1,
placeholder="describe your video content"
# value="bear, Van Gogh Style"
)
sr_ratio = gr.Slider(label='SR ratio',
minimum=1,
maximum=16,
value=4,
step=1)
n_frames = gr.Slider(label='Frames',
minimum=1,
maximum=60,
value=10,
step=1)
n_steps = gr.Slider(label='Steps',
minimum=1,
maximum=100,
value=10,
step=1)
guidance_scale = gr.Slider(label='Guidance Scale',
minimum=0.1,
maximum=30.0,
value=4.0,
step=0.1)
seed = gr.Slider(label='Seed',
minimum=-1,
maximum=1000,
step=1,
randomize=True)
n_prompt = gr.Textbox(
label='Negative Prompt',
value="low quality, blurry, low-resolution, noisy, unsharp, weird textures"
)
task = gr.Textbox(value="sr", visible=False)
# input_video.change(
# fn = update_prompt,
# inputs = [input_video],
# outputs = [prompt],
# queue = False)
run_button.click(fn = DiffBIR_restore,
inputs = [input_video,
prompt,
sr_ratio,
n_frames,
n_steps,
guidance_scale,
seed,
n_prompt,
task
],
outputs = [output_video]
)
gr.Examples(
examples=get_example("sr"),
label='Examples',
inputs=[input_video],
outputs=[output_video],
examples_per_page=7
)
with gr.Tab(label="Denoise with DiffBIR"):
with gr.Row():
input_video = gr.Video(label="Input Video")
output_video = gr.Video(label="Restored Video", interactive=False)
with gr.Row():
run_button = gr.Button("Restore your video !", visible=True)
with gr.Accordion('Advanced options', open=False):
prompt = gr.Textbox(
label="Prompt",
max_lines=1,
placeholder="describe your video content"
# value="bear, Van Gogh Style"
)
n_frames = gr.Slider(label='Frames',
minimum=1,
maximum=60,
value=10,
step=1)
n_steps = gr.Slider(label='Steps',
minimum=1,
maximum=100,
value=10,
step=1)
guidance_scale = gr.Slider(label='Guidance Scale',
minimum=0.1,
maximum=30.0,
value=4.0,
step=0.1)
seed = gr.Slider(label='Seed',
minimum=-1,
maximum=1000,
step=1,
randomize=True)
n_prompt = gr.Textbox(
label='Negative Prompt',
value="low quality, blurry, low-resolution, noisy, unsharp, weird textures"
)
task = gr.Textbox(value="dn", visible=False)
sr_ratio = gr.Number(value=1, visible=False)
# input_video.change(
# fn = update_prompt,
# inputs = [input_video],
# outputs = [prompt],
# queue = False)
run_button.click(fn = DiffBIR_restore,
inputs = [input_video,
prompt,
sr_ratio,
n_frames,
n_steps,
guidance_scale,
seed,
n_prompt,
task
],
outputs = [output_video]
)
gr.Examples(
examples=get_example("dn"),
label='Examples',
inputs=[input_video],
outputs=[output_video],
examples_per_page=7
)
demo.queue()
demo.launch(share=True) |