Spaces:
Running
on
A10G
Running
on
A10G
Shanshan Wang
commited on
Commit
•
471e971
1
Parent(s):
b596e22
updated app
Browse files- app.py +235 -4
- requirements.txt +9 -0
app.py
CHANGED
@@ -1,7 +1,238 @@
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
return "Hello " + name + "!!"
|
5 |
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from transformers import AutoModel, AutoTokenizer, AutoImageProcessor
|
4 |
+
import torch
|
5 |
+
import torchvision.transforms as T
|
6 |
+
from PIL import Image
|
7 |
+
import time
|
8 |
+
import os, sys
|
9 |
+
import json
|
10 |
+
import re
|
11 |
+
from tqdm import tqdm
|
12 |
|
13 |
+
import pandas as pd
|
|
|
14 |
|
15 |
+
from torchvision.transforms.functional import InterpolationMode
|
16 |
+
# Define the path to your model
|
17 |
+
path = 'h2oai/h2o-mississippi-2b'
|
18 |
+
|
19 |
+
# image preprocesing
|
20 |
+
IMAGENET_MEAN = (0.485, 0.456, 0.406)
|
21 |
+
IMAGENET_STD = (0.229, 0.224, 0.225)
|
22 |
+
|
23 |
+
start_pre = time.time()
|
24 |
+
|
25 |
+
def build_transform(input_size):
|
26 |
+
MEAN, STD = IMAGENET_MEAN, IMAGENET_STD
|
27 |
+
transform = T.Compose([
|
28 |
+
T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img),
|
29 |
+
T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC),
|
30 |
+
T.ToTensor(),
|
31 |
+
T.Normalize(mean=MEAN, std=STD)
|
32 |
+
])
|
33 |
+
return transform
|
34 |
+
|
35 |
+
|
36 |
+
def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
|
37 |
+
best_ratio_diff = float('inf')
|
38 |
+
best_ratio = (1, 1)
|
39 |
+
area = width * height
|
40 |
+
for ratio in target_ratios:
|
41 |
+
target_aspect_ratio = ratio[0] / ratio[1]
|
42 |
+
ratio_diff = abs(aspect_ratio - target_aspect_ratio)
|
43 |
+
if ratio_diff < best_ratio_diff:
|
44 |
+
best_ratio_diff = ratio_diff
|
45 |
+
best_ratio = ratio
|
46 |
+
elif ratio_diff == best_ratio_diff:
|
47 |
+
if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]:
|
48 |
+
best_ratio = ratio
|
49 |
+
return best_ratio
|
50 |
+
def dynamic_preprocess(image, min_num=1, max_num=6, image_size=448, use_thumbnail=False):
|
51 |
+
orig_width, orig_height = image.size
|
52 |
+
aspect_ratio = orig_width / orig_height
|
53 |
+
|
54 |
+
# calculate the existing image aspect ratio
|
55 |
+
target_ratios = set(
|
56 |
+
(i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if
|
57 |
+
i * j <= max_num and i * j >= min_num)
|
58 |
+
target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])
|
59 |
+
|
60 |
+
# find the closest aspect ratio to the target
|
61 |
+
target_aspect_ratio = find_closest_aspect_ratio(
|
62 |
+
aspect_ratio, target_ratios, orig_width, orig_height, image_size)
|
63 |
+
|
64 |
+
# calculate the target width and height
|
65 |
+
target_width = image_size * target_aspect_ratio[0]
|
66 |
+
target_height = image_size * target_aspect_ratio[1]
|
67 |
+
blocks = target_aspect_ratio[0] * target_aspect_ratio[1]
|
68 |
+
|
69 |
+
# resize the image
|
70 |
+
resized_img = image.resize((target_width, target_height))
|
71 |
+
processed_images = []
|
72 |
+
for i in range(blocks):
|
73 |
+
box = (
|
74 |
+
(i % (target_width // image_size)) * image_size,
|
75 |
+
(i // (target_width // image_size)) * image_size,
|
76 |
+
((i % (target_width // image_size)) + 1) * image_size,
|
77 |
+
((i // (target_width // image_size)) + 1) * image_size
|
78 |
+
)
|
79 |
+
# split the image
|
80 |
+
split_img = resized_img.crop(box)
|
81 |
+
processed_images.append(split_img)
|
82 |
+
assert len(processed_images) == blocks
|
83 |
+
if use_thumbnail and len(processed_images) != 1:
|
84 |
+
thumbnail_img = image.resize((image_size, image_size))
|
85 |
+
processed_images.append(thumbnail_img)
|
86 |
+
return processed_images, target_aspect_ratio
|
87 |
+
|
88 |
+
|
89 |
+
def dynamic_preprocess2(image, min_num=1, max_num=6, image_size=448, use_thumbnail=False, prior_aspect_ratio=None):
|
90 |
+
orig_width, orig_height = image.size
|
91 |
+
aspect_ratio = orig_width / orig_height
|
92 |
+
|
93 |
+
# calculate the existing image aspect ratio
|
94 |
+
target_ratios = set(
|
95 |
+
(i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if
|
96 |
+
i * j <= max_num and i * j >= min_num)
|
97 |
+
target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])
|
98 |
+
|
99 |
+
new_target_ratios = []
|
100 |
+
if prior_aspect_ratio is not None:
|
101 |
+
for i in target_ratios:
|
102 |
+
if prior_aspect_ratio[0]%i[0] != 0 and prior_aspect_ratio[1]%i[1] != 0:
|
103 |
+
new_target_ratios.append(i)
|
104 |
+
else:
|
105 |
+
continue
|
106 |
+
|
107 |
+
# find the closest aspect ratio to the target
|
108 |
+
target_aspect_ratio = find_closest_aspect_ratio(
|
109 |
+
aspect_ratio, new_target_ratios, orig_width, orig_height, image_size)
|
110 |
+
|
111 |
+
# calculate the target width and height
|
112 |
+
target_width = image_size * target_aspect_ratio[0]
|
113 |
+
target_height = image_size * target_aspect_ratio[1]
|
114 |
+
blocks = target_aspect_ratio[0] * target_aspect_ratio[1]
|
115 |
+
|
116 |
+
# resize the image
|
117 |
+
resized_img = image.resize((target_width, target_height))
|
118 |
+
processed_images = []
|
119 |
+
for i in range(blocks):
|
120 |
+
box = (
|
121 |
+
(i % (target_width // image_size)) * image_size,
|
122 |
+
(i // (target_width // image_size)) * image_size,
|
123 |
+
((i % (target_width // image_size)) + 1) * image_size,
|
124 |
+
((i // (target_width // image_size)) + 1) * image_size
|
125 |
+
)
|
126 |
+
# split the image
|
127 |
+
split_img = resized_img.crop(box)
|
128 |
+
processed_images.append(split_img)
|
129 |
+
assert len(processed_images) == blocks
|
130 |
+
if use_thumbnail and len(processed_images) != 1:
|
131 |
+
thumbnail_img = image.resize((image_size, image_size))
|
132 |
+
processed_images.append(thumbnail_img)
|
133 |
+
return processed_images
|
134 |
+
def load_image1(image_file, input_size=448, min_num=1, max_num=12):
|
135 |
+
if isinstance(image_file, str):
|
136 |
+
image = Image.open(image_file).convert('RGB')
|
137 |
+
else:
|
138 |
+
image = image_file
|
139 |
+
transform = build_transform(input_size=input_size)
|
140 |
+
images, target_aspect_ratio = dynamic_preprocess(image, image_size=input_size, use_thumbnail=True, min_num=min_num, max_num=max_num)
|
141 |
+
pixel_values = [transform(image) for image in images]
|
142 |
+
pixel_values = torch.stack(pixel_values)
|
143 |
+
return pixel_values, target_aspect_ratio
|
144 |
+
|
145 |
+
def load_image2(image_file, input_size=448, min_num=1, max_num=12, target_aspect_ratio=None):
|
146 |
+
|
147 |
+
if isinstance(image_file, str):
|
148 |
+
image = Image.open(image_file).convert('RGB')
|
149 |
+
else:
|
150 |
+
image = image_file
|
151 |
+
transform = build_transform(input_size=input_size)
|
152 |
+
images = dynamic_preprocess2(image, image_size=input_size, use_thumbnail=True, min_num=min_num, max_num=max_num, prior_aspect_ratio=target_aspect_ratio)
|
153 |
+
pixel_values = [transform(image) for image in images]
|
154 |
+
pixel_values = torch.stack(pixel_values)
|
155 |
+
return pixel_values
|
156 |
+
|
157 |
+
def load_image_msac(file_name):
|
158 |
+
pixel_values, target_aspect_ratio = load_image1(file_name, min_num=1, max_num=6)
|
159 |
+
pixel_values = pixel_values.to(torch.bfloat16).cuda()
|
160 |
+
pixel_values2 = load_image2(file_name, min_num=3, max_num=6, target_aspect_ratio=target_aspect_ratio)
|
161 |
+
pixel_values2 = pixel_values2.to(torch.bfloat16).cuda()
|
162 |
+
pixel_values = torch.cat([pixel_values2[:-1], pixel_values[:-1], pixel_values2[-1:]], 0)
|
163 |
+
return pixel_values
|
164 |
+
# Load the model and tokenizer
|
165 |
+
model = AutoModel.from_pretrained(
|
166 |
+
path,
|
167 |
+
torch_dtype=torch.bfloat16,
|
168 |
+
low_cpu_mem_usage=True,
|
169 |
+
trust_remote_code=True
|
170 |
+
).eval().cuda()
|
171 |
+
|
172 |
+
tokenizer = AutoTokenizer.from_pretrained(
|
173 |
+
path,
|
174 |
+
trust_remote_code=True,
|
175 |
+
use_fast=False
|
176 |
+
)
|
177 |
+
tokenizer.pad_token = tokenizer.unk_token
|
178 |
+
tokenizer.eos_token = "<|end|>"
|
179 |
+
model.generation_config.pad_token_id = tokenizer.pad_token_id
|
180 |
+
|
181 |
+
|
182 |
+
def inference(image, prompt):
|
183 |
+
# Check if both image and prompt are provided
|
184 |
+
if image is None or prompt.strip() == "":
|
185 |
+
return "Please provide both an image and a prompt."
|
186 |
+
|
187 |
+
# Process the image and get pixel_values
|
188 |
+
pixel_values = load_image_msac(image)
|
189 |
+
|
190 |
+
# Set generation config
|
191 |
+
generation_config = dict(
|
192 |
+
num_beams=1,
|
193 |
+
max_new_tokens=2048,
|
194 |
+
do_sample=False,
|
195 |
+
)
|
196 |
+
|
197 |
+
# Generate the response
|
198 |
+
response = model.chat(
|
199 |
+
tokenizer,
|
200 |
+
pixel_values,
|
201 |
+
prompt,
|
202 |
+
generation_config
|
203 |
+
)
|
204 |
+
|
205 |
+
return response
|
206 |
+
|
207 |
+
# Build the Gradio interface
|
208 |
+
with gr.Blocks() as demo:
|
209 |
+
gr.Markdown("H2O-Mississippi")
|
210 |
+
|
211 |
+
with gr.Row():
|
212 |
+
image_input = gr.Image(type="pil", label="Upload an Image")
|
213 |
+
prompt_input = gr.Textbox(label="Enter your prompt here")
|
214 |
+
|
215 |
+
response_output = gr.Textbox(label="Model Response")
|
216 |
+
|
217 |
+
with gr.Row():
|
218 |
+
submit_button = gr.Button("Submit")
|
219 |
+
clear_button = gr.Button("Clear")
|
220 |
+
|
221 |
+
# When the submit button is clicked, call the inference function
|
222 |
+
submit_button.click(
|
223 |
+
fn=inference,
|
224 |
+
inputs=[image_input, prompt_input],
|
225 |
+
outputs=response_output
|
226 |
+
)
|
227 |
+
|
228 |
+
# Define the clear button action
|
229 |
+
def clear_all():
|
230 |
+
return None, "", ""
|
231 |
+
|
232 |
+
clear_button.click(
|
233 |
+
fn=clear_all,
|
234 |
+
inputs=None,
|
235 |
+
outputs=[image_input, prompt_input, response_output]
|
236 |
+
)
|
237 |
+
|
238 |
+
demo.launch(share=True)
|
requirements.txt
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
fastapi
|
2 |
+
opencv-python
|
3 |
+
gradio==3.35.2
|
4 |
+
gradio_client==0.2.9
|
5 |
+
httpx==0.24.0
|
6 |
+
markdown2[all]
|
7 |
+
pydantic
|
8 |
+
requests
|
9 |
+
uvicorn
|