Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,225 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
from unittest.mock import patch
|
3 |
+
import spaces
|
4 |
+
import gradio as gr
|
5 |
+
from transformers import AutoProcessor, AutoModelForCausalLM
|
6 |
+
from transformers.dynamic_module_utils import get_imports
|
7 |
+
import torch
|
8 |
+
import requests
|
9 |
+
from PIL import Image, ImageDraw
|
10 |
+
import random
|
11 |
+
import numpy as np
|
12 |
+
import matplotlib.pyplot as plt
|
13 |
+
import matplotlib.patches as patches
|
14 |
+
import cv2
|
15 |
+
import io
|
16 |
+
|
17 |
+
def workaround_fixed_get_imports(filename: str | os.PathLike) -> list[str]:
|
18 |
+
if not str(filename).endswith("/modeling_florence2.py"):
|
19 |
+
return get_imports(filename)
|
20 |
+
imports = get_imports(filename)
|
21 |
+
imports.remove("flash_attn")
|
22 |
+
return imports
|
23 |
+
|
24 |
+
with patch("transformers.dynamic_module_utils.get_imports", workaround_fixed_get_imports):
|
25 |
+
model = AutoModelForCausalLM.from_pretrained("microsoft/Florence-2-large-ft", trust_remote_code=True).to("cuda").eval()
|
26 |
+
processor = AutoProcessor.from_pretrained("microsoft/Florence-2-large-ft", trust_remote_code=True)
|
27 |
+
|
28 |
+
colormap = ['blue', 'orange', 'green', 'purple', 'brown', 'pink', 'gray', 'olive', 'cyan', 'red',
|
29 |
+
'lime', 'indigo', 'violet', 'aqua', 'magenta', 'coral', 'gold', 'tan', 'skyblue']
|
30 |
+
|
31 |
+
def fig_to_pil(fig):
|
32 |
+
buf = io.BytesIO()
|
33 |
+
fig.savefig(buf, format='png')
|
34 |
+
buf.seek(0)
|
35 |
+
return Image.open(buf)
|
36 |
+
|
37 |
+
@spaces.GPU
|
38 |
+
def run_example(task_prompt, image, text_input=None):
|
39 |
+
if text_input is None:
|
40 |
+
prompt = task_prompt
|
41 |
+
else:
|
42 |
+
prompt = task_prompt + text_input
|
43 |
+
inputs = processor(text=prompt, images=image, return_tensors="pt").to("cuda")
|
44 |
+
with torch.inference_mode():
|
45 |
+
generated_ids = model.generate(
|
46 |
+
input_ids=inputs["input_ids"],
|
47 |
+
pixel_values=inputs["pixel_values"],
|
48 |
+
max_new_tokens=1024,
|
49 |
+
early_stopping=False,
|
50 |
+
do_sample=False,
|
51 |
+
num_beams=3,
|
52 |
+
)
|
53 |
+
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
|
54 |
+
parsed_answer = processor.post_process_generation(
|
55 |
+
generated_text,
|
56 |
+
task=task_prompt,
|
57 |
+
image_size=(image.size[0], image.size[1])
|
58 |
+
)
|
59 |
+
return parsed_answer
|
60 |
+
|
61 |
+
def plot_bbox(image, data):
|
62 |
+
fig, ax = plt.subplots()
|
63 |
+
ax.imshow(image)
|
64 |
+
for bbox, label in zip(data['bboxes'], data['labels']):
|
65 |
+
x1, y1, x2, y2 = bbox
|
66 |
+
rect = patches.Rectangle((x1, y1), x2-x1, y2-y1, linewidth=1, edgecolor='r', facecolor='none')
|
67 |
+
ax.add_patch(rect)
|
68 |
+
plt.text(x1, y1, label, color='white', fontsize=8, bbox=dict(facecolor='indigo', alpha=0.5))
|
69 |
+
ax.axis('off')
|
70 |
+
return fig_to_pil(fig)
|
71 |
+
|
72 |
+
def draw_polygons(image, prediction, fill_mask=False):
|
73 |
+
fig, ax = plt.subplots()
|
74 |
+
ax.imshow(image)
|
75 |
+
scale = 1
|
76 |
+
for polygons, label in zip(prediction['polygons'], prediction['labels']):
|
77 |
+
color = random.choice(colormap)
|
78 |
+
fill_color = random.choice(colormap) if fill_mask else None
|
79 |
+
for _polygon in polygons:
|
80 |
+
_polygon = np.array(_polygon).reshape(-1, 2)
|
81 |
+
if _polygon.shape[0] < 3:
|
82 |
+
print('Invalid polygon:', _polygon)
|
83 |
+
continue
|
84 |
+
_polygon = (_polygon * scale).reshape(-1).tolist()
|
85 |
+
if len(_polygon) % 2 != 0:
|
86 |
+
print('Invalid polygon:', _polygon)
|
87 |
+
continue
|
88 |
+
polygon_points = np.array(_polygon).reshape(-1, 2)
|
89 |
+
if fill_mask:
|
90 |
+
polygon = patches.Polygon(polygon_points, edgecolor=color, facecolor=fill_color, linewidth=2)
|
91 |
+
else:
|
92 |
+
polygon = patches.Polygon(polygon_points, edgecolor=color, fill=False, linewidth=2)
|
93 |
+
ax.add_patch(polygon)
|
94 |
+
plt.text(polygon_points[0, 0], polygon_points[0, 1], label, color='white', fontsize=8, bbox=dict(facecolor=color, alpha=0.5))
|
95 |
+
ax.axis('off')
|
96 |
+
return fig_to_pil(fig)
|
97 |
+
|
98 |
+
def draw_ocr_bboxes(image, prediction):
|
99 |
+
fig, ax = plt.subplots()
|
100 |
+
ax.imshow(image)
|
101 |
+
scale = 1
|
102 |
+
bboxes, labels = prediction['quad_boxes'], prediction['labels']
|
103 |
+
for box, label in zip(bboxes, labels):
|
104 |
+
color = random.choice(colormap)
|
105 |
+
new_box = (np.array(box) * scale).tolist()
|
106 |
+
polygon = patches.Polygon(new_box, edgecolor=color, fill=False, linewidth=3)
|
107 |
+
ax.add_patch(polygon)
|
108 |
+
plt.text(new_box[0], new_box[1], label, color='white', fontsize=8, bbox=dict(facecolor=color, alpha=0.5))
|
109 |
+
ax.axis('off')
|
110 |
+
return fig_to_pil(fig)
|
111 |
+
|
112 |
+
|
113 |
+
@spaces.GPU(duration=120)
|
114 |
+
def process_video(input_video_path, task_prompt):
|
115 |
+
cap = cv2.VideoCapture(input_video_path)
|
116 |
+
if not cap.isOpened():
|
117 |
+
print("Error: Can't open the video file.")
|
118 |
+
return
|
119 |
+
|
120 |
+
frame_width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
|
121 |
+
frame_height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
|
122 |
+
fps = cap.get(cv2.CAP_PROP_FPS)
|
123 |
+
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
|
124 |
+
out = cv2.VideoWriter("output_vid.mp4", fourcc, fps, (frame_width, frame_height))
|
125 |
+
|
126 |
+
while cap.isOpened():
|
127 |
+
ret, frame = cap.read()
|
128 |
+
if not ret:
|
129 |
+
break
|
130 |
+
|
131 |
+
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
|
132 |
+
pil_image = Image.fromarray(frame_rgb)
|
133 |
+
|
134 |
+
result = run_example(task_prompt, pil_image)
|
135 |
+
|
136 |
+
if task_prompt == "<OD>":
|
137 |
+
processed_image = plot_bbox(pil_image, result['<OD>'])
|
138 |
+
elif task_prompt == "<DENSE_REGION_CAPTION>":
|
139 |
+
processed_image = plot_bbox(pil_image, result['<DENSE_REGION_CAPTION>'])
|
140 |
+
else:
|
141 |
+
processed_image = pil_image
|
142 |
+
|
143 |
+
processed_frame = cv2.cvtColor(np.array(processed_image), cv2.COLOR_RGB2BGR)
|
144 |
+
out.write(processed_frame)
|
145 |
+
|
146 |
+
cap.release()
|
147 |
+
out.release()
|
148 |
+
cv2.destroyAllWindows()
|
149 |
+
return "output_vid.mp4"
|
150 |
+
|
151 |
+
css = """
|
152 |
+
#output {
|
153 |
+
min-height: 100px;
|
154 |
+
overflow: auto;
|
155 |
+
border: 1px solid #ccc;
|
156 |
+
}
|
157 |
+
"""
|
158 |
+
|
159 |
+
with gr.Blocks(css=css) as demo:
|
160 |
+
gr.HTML("<h1><center>Microsoft Florence-2-large-ft</center></h1>")
|
161 |
+
with gr.Tab(label="Image"):
|
162 |
+
with gr.Row():
|
163 |
+
with gr.Column():
|
164 |
+
input_img = gr.Image(label="Input Picture", type="pil")
|
165 |
+
task_radio = gr.Radio(
|
166 |
+
["Caption", "Detailed Caption", "More Detailed Caption", "Caption to Phrase Grounding",
|
167 |
+
"Object Detection", "Dense Region Caption", "Region Proposal", "Referring Expression Segmentation",
|
168 |
+
"Region to Segmentation", "Open Vocabulary Detection", "Region to Category", "Region to Description",
|
169 |
+
"OCR", "OCR with Region"],
|
170 |
+
label="Task", value="Caption"
|
171 |
+
)
|
172 |
+
text_input = gr.Textbox(label="Text Input (is Optional)", visible=False)
|
173 |
+
submit_btn = gr.Button(value="Submit")
|
174 |
+
with gr.Column():
|
175 |
+
output_text = gr.Textbox(label="Results")
|
176 |
+
output_image = gr.Image(label="Image", type="pil")
|
177 |
+
|
178 |
+
with gr.Tab(label="Video"):
|
179 |
+
with gr.Row():
|
180 |
+
with gr.Column():
|
181 |
+
input_video = gr.Video(label="Video")
|
182 |
+
video_task_radio = gr.Radio(
|
183 |
+
["Object Detection", "Dense Region Caption"],
|
184 |
+
label="Video Task", value="Object Detection"
|
185 |
+
)
|
186 |
+
video_submit_btn = gr.Button(value="Process Video")
|
187 |
+
with gr.Column():
|
188 |
+
output_video = gr.Video(label="Video")
|
189 |
+
|
190 |
+
def update_text_input(task):
|
191 |
+
return gr.update(visible=task in ["Caption to Phrase Grounding", "Referring Expression Segmentation",
|
192 |
+
"Region to Segmentation", "Open Vocabulary Detection", "Region to Category",
|
193 |
+
"Region to Description"])
|
194 |
+
|
195 |
+
task_radio.change(fn=update_text_input, inputs=task_radio, outputs=text_input)
|
196 |
+
|
197 |
+
def process_image(image, task, text):
|
198 |
+
task_mapping = {
|
199 |
+
"Caption": ("<CAPTION>", lambda result: (result['<CAPTION>'], image)),
|
200 |
+
"Detailed Caption": ("<DETAILED_CAPTION>", lambda result: (result['<DETAILED_CAPTION>'], image)),
|
201 |
+
"More Detailed Caption": ("<MORE_DETAILED_CAPTION>", lambda result: (result['<MORE_DETAILED_CAPTION>'], image)),
|
202 |
+
"Caption to Phrase Grounding": ("<CAPTION_TO_PHRASE_GROUNDING>", lambda result: (str(result['<CAPTION_TO_PHRASE_GROUNDING>']), plot_bbox(image, result['<CAPTION_TO_PHRASE_GROUNDING>']))),
|
203 |
+
"Object Detection": ("<OD>", lambda result: (str(result['<OD>']), plot_bbox(image, result['<OD>']))),
|
204 |
+
"Dense Region Caption": ("<DENSE_REGION_CAPTION>", lambda result: (str(result['<DENSE_REGION_CAPTION>']), plot_bbox(image, result['<DENSE_REGION_CAPTION>']))),
|
205 |
+
"Region Proposal": ("<REGION_PROPOSAL>", lambda result: (str(result['<REGION_PROPOSAL>']), plot_bbox(image, result['<REGION_PROPOSAL>']))),
|
206 |
+
"Referring Expression Segmentation": ("<REFERRING_EXPRESSION_SEGMENTATION>", lambda result: (str(result['<REFERRING_EXPRESSION_SEGMENTATION>']), draw_polygons(image, result['<REFERRING_EXPRESSION_SEGMENTATION>'], fill_mask=True))),
|
207 |
+
"Region to Segmentation": ("<REGION_TO_SEGMENTATION>", lambda result: (str(result['<REGION_TO_SEGMENTATION>']), draw_polygons(image, result['<REGION_TO_SEGMENTATION>'], fill_mask=True))),
|
208 |
+
"Open Vocabulary Detection": ("<OPEN_VOCABULARY_DETECTION>", lambda result: (str(convert_to_od_format(result['<OPEN_VOCABULARY_DETECTION>'])), plot_bbox(image, convert_to_od_format(result['<OPEN_VOCABULARY_DETECTION>'])))),
|
209 |
+
"Region to Category": ("<REGION_TO_CATEGORY>", lambda result: (result['<REGION_TO_CATEGORY>'], image)),
|
210 |
+
"Region to Description": ("<REGION_TO_DESCRIPTION>", lambda result: (result['<REGION_TO_DESCRIPTION>'], image)),
|
211 |
+
"OCR": ("<OCR>", lambda result: (result['<OCR>'], image)),
|
212 |
+
"OCR with Region": ("<OCR_WITH_REGION>", lambda result: (str(result['<OCR_WITH_REGION>']), draw_ocr_bboxes(image, result['<OCR_WITH_REGION>']))),
|
213 |
+
}
|
214 |
+
|
215 |
+
if task in task_mapping:
|
216 |
+
prompt, process_func = task_mapping[task]
|
217 |
+
result = run_example(prompt, image, text)
|
218 |
+
return process_func(result)
|
219 |
+
else:
|
220 |
+
return "", image
|
221 |
+
|
222 |
+
submit_btn.click(fn=process_image, inputs=[input_img, task_radio, text_input], outputs=[output_text, output_image])
|
223 |
+
video_submit_btn.click(fn=process_video, inputs=[input_video, video_task_radio], outputs=output_video)
|
224 |
+
|
225 |
+
demo.launch()
|