Spaces:
Running
Running
Commit
β’
e25fb5a
1
Parent(s):
e5535b9
Upload 6 files
Browse files- README.md +4 -4
- app.py +233 -0
- image1.jpg +0 -0
- image2.jpg +0 -0
- pre-requirements.txt +1 -0
- requirements.txt +3 -0
README.md
CHANGED
@@ -1,8 +1,8 @@
|
|
1 |
---
|
2 |
-
title: Florence
|
3 |
-
emoji:
|
4 |
-
colorFrom:
|
5 |
-
colorTo:
|
6 |
sdk: gradio
|
7 |
sdk_version: 4.36.1
|
8 |
app_file: app.py
|
|
|
1 |
---
|
2 |
+
title: Florence-2-base
|
3 |
+
emoji: π
|
4 |
+
colorFrom: pink
|
5 |
+
colorTo: purple
|
6 |
sdk: gradio
|
7 |
sdk_version: 4.36.1
|
8 |
app_file: app.py
|
app.py
ADDED
@@ -0,0 +1,233 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from transformers import AutoProcessor, AutoModelForCausalLM
|
3 |
+
#import spaces
|
4 |
+
|
5 |
+
import requests
|
6 |
+
import copy
|
7 |
+
|
8 |
+
from PIL import Image, ImageDraw, ImageFont
|
9 |
+
import io
|
10 |
+
import matplotlib.pyplot as plt
|
11 |
+
import matplotlib.patches as patches
|
12 |
+
|
13 |
+
import random
|
14 |
+
import numpy as np
|
15 |
+
|
16 |
+
import subprocess
|
17 |
+
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
|
18 |
+
|
19 |
+
models = {
|
20 |
+
'microsoft/Florence-2-base': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True).to("cuda").eval(),
|
21 |
+
}
|
22 |
+
|
23 |
+
processors = {
|
24 |
+
'microsoft/Florence-2-base': AutoProcessor.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True),
|
25 |
+
}
|
26 |
+
|
27 |
+
|
28 |
+
DESCRIPTION = "# [Florence-2 Demo](https://huggingface.co/microsoft/Florence-2-base)"
|
29 |
+
|
30 |
+
colormap = ['blue','orange','green','purple','brown','pink','gray','olive','cyan','red',
|
31 |
+
'lime','indigo','violet','aqua','magenta','coral','gold','tan','skyblue']
|
32 |
+
|
33 |
+
def fig_to_pil(fig):
|
34 |
+
buf = io.BytesIO()
|
35 |
+
fig.savefig(buf, format='png')
|
36 |
+
buf.seek(0)
|
37 |
+
return Image.open(buf)
|
38 |
+
|
39 |
+
#@spaces.GPU
|
40 |
+
def run_example(task_prompt, image, text_input=None, model_id='microsoft/Florence-2-base'):
|
41 |
+
model = models[model_id]
|
42 |
+
processor = processors[model_id]
|
43 |
+
if text_input is None:
|
44 |
+
prompt = task_prompt
|
45 |
+
else:
|
46 |
+
prompt = task_prompt + text_input
|
47 |
+
inputs = processor(text=prompt, images=image, return_tensors="pt").to("cuda")
|
48 |
+
generated_ids = model.generate(
|
49 |
+
input_ids=inputs["input_ids"],
|
50 |
+
pixel_values=inputs["pixel_values"],
|
51 |
+
max_new_tokens=1024,
|
52 |
+
early_stopping=False,
|
53 |
+
do_sample=False,
|
54 |
+
num_beams=3,
|
55 |
+
)
|
56 |
+
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
|
57 |
+
parsed_answer = processor.post_process_generation(
|
58 |
+
generated_text,
|
59 |
+
task=task_prompt,
|
60 |
+
image_size=(image.width, image.height)
|
61 |
+
)
|
62 |
+
return parsed_answer
|
63 |
+
|
64 |
+
def plot_bbox(image, data):
|
65 |
+
fig, ax = plt.subplots()
|
66 |
+
ax.imshow(image)
|
67 |
+
for bbox, label in zip(data['bboxes'], data['labels']):
|
68 |
+
x1, y1, x2, y2 = bbox
|
69 |
+
rect = patches.Rectangle((x1, y1), x2-x1, y2-y1, linewidth=1, edgecolor='r', facecolor='none')
|
70 |
+
ax.add_patch(rect)
|
71 |
+
plt.text(x1, y1, label, color='white', fontsize=8, bbox=dict(facecolor='red', alpha=0.5))
|
72 |
+
ax.axis('off')
|
73 |
+
return fig
|
74 |
+
|
75 |
+
def draw_polygons(image, prediction, fill_mask=False):
|
76 |
+
|
77 |
+
draw = ImageDraw.Draw(image)
|
78 |
+
scale = 1
|
79 |
+
for polygons, label in zip(prediction['polygons'], prediction['labels']):
|
80 |
+
color = random.choice(colormap)
|
81 |
+
fill_color = random.choice(colormap) if fill_mask else None
|
82 |
+
for _polygon in polygons:
|
83 |
+
_polygon = np.array(_polygon).reshape(-1, 2)
|
84 |
+
if len(_polygon) < 3:
|
85 |
+
print('Invalid polygon:', _polygon)
|
86 |
+
continue
|
87 |
+
_polygon = (_polygon * scale).reshape(-1).tolist()
|
88 |
+
if fill_mask:
|
89 |
+
draw.polygon(_polygon, outline=color, fill=fill_color)
|
90 |
+
else:
|
91 |
+
draw.polygon(_polygon, outline=color)
|
92 |
+
draw.text((_polygon[0] + 8, _polygon[1] + 2), label, fill=color)
|
93 |
+
return image
|
94 |
+
|
95 |
+
def convert_to_od_format(data):
|
96 |
+
bboxes = data.get('bboxes', [])
|
97 |
+
labels = data.get('bboxes_labels', [])
|
98 |
+
od_results = {
|
99 |
+
'bboxes': bboxes,
|
100 |
+
'labels': labels
|
101 |
+
}
|
102 |
+
return od_results
|
103 |
+
|
104 |
+
def draw_ocr_bboxes(image, prediction):
|
105 |
+
scale = 1
|
106 |
+
draw = ImageDraw.Draw(image)
|
107 |
+
bboxes, labels = prediction['quad_boxes'], prediction['labels']
|
108 |
+
for box, label in zip(bboxes, labels):
|
109 |
+
color = random.choice(colormap)
|
110 |
+
new_box = (np.array(box) * scale).tolist()
|
111 |
+
draw.polygon(new_box, width=3, outline=color)
|
112 |
+
draw.text((new_box[0]+8, new_box[1]+2),
|
113 |
+
"{}".format(label),
|
114 |
+
align="right",
|
115 |
+
fill=color)
|
116 |
+
return image
|
117 |
+
|
118 |
+
def process_image(image, task_prompt, text_input=None, model_id='microsoft/Florence-2-base'):
|
119 |
+
image = Image.fromarray(image) # Convert NumPy array to PIL Image
|
120 |
+
if task_prompt == 'Caption':
|
121 |
+
task_prompt = '<CAPTION>'
|
122 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
123 |
+
return results, None
|
124 |
+
elif task_prompt == 'Detailed Caption':
|
125 |
+
task_prompt = '<DETAILED_CAPTION>'
|
126 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
127 |
+
return results, None
|
128 |
+
elif task_prompt == 'More Detailed Caption':
|
129 |
+
task_prompt = '<MORE_DETAILED_CAPTION>'
|
130 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
131 |
+
return results, None
|
132 |
+
elif task_prompt == 'Object Detection':
|
133 |
+
task_prompt = '<OD>'
|
134 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
135 |
+
fig = plot_bbox(image, results['<OD>'])
|
136 |
+
return results, fig_to_pil(fig)
|
137 |
+
elif task_prompt == 'Dense Region Caption':
|
138 |
+
task_prompt = '<DENSE_REGION_CAPTION>'
|
139 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
140 |
+
fig = plot_bbox(image, results['<DENSE_REGION_CAPTION>'])
|
141 |
+
return results, fig_to_pil(fig)
|
142 |
+
elif task_prompt == 'Region Proposal':
|
143 |
+
task_prompt = '<REGION_PROPOSAL>'
|
144 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
145 |
+
fig = plot_bbox(image, results['<REGION_PROPOSAL>'])
|
146 |
+
return results, fig_to_pil(fig)
|
147 |
+
elif task_prompt == 'Caption to Phrase Grounding':
|
148 |
+
task_prompt = '<CAPTION_TO_PHRASE_GROUNDING>'
|
149 |
+
results = run_example(task_prompt, image, text_input, model_id)
|
150 |
+
fig = plot_bbox(image, results['<CAPTION_TO_PHRASE_GROUNDING>'])
|
151 |
+
return results, fig_to_pil(fig)
|
152 |
+
elif task_prompt == 'Referring Expression Segmentation':
|
153 |
+
task_prompt = '<REFERRING_EXPRESSION_SEGMENTATION>'
|
154 |
+
results = run_example(task_prompt, image, text_input, model_id)
|
155 |
+
output_image = copy.deepcopy(image)
|
156 |
+
output_image = draw_polygons(output_image, results['<REFERRING_EXPRESSION_SEGMENTATION>'], fill_mask=True)
|
157 |
+
return results, output_image
|
158 |
+
elif task_prompt == 'Region to Segmentation':
|
159 |
+
task_prompt = '<REGION_TO_SEGMENTATION>'
|
160 |
+
results = run_example(task_prompt, image, text_input, model_id)
|
161 |
+
output_image = copy.deepcopy(image)
|
162 |
+
output_image = draw_polygons(output_image, results['<REGION_TO_SEGMENTATION>'], fill_mask=True)
|
163 |
+
return results, output_image
|
164 |
+
elif task_prompt == 'Open Vocabulary Detection':
|
165 |
+
task_prompt = '<OPEN_VOCABULARY_DETECTION>'
|
166 |
+
results = run_example(task_prompt, image, text_input, model_id)
|
167 |
+
bbox_results = convert_to_od_format(results['<OPEN_VOCABULARY_DETECTION>'])
|
168 |
+
fig = plot_bbox(image, bbox_results)
|
169 |
+
return results, fig_to_pil(fig)
|
170 |
+
elif task_prompt == 'Region to Category':
|
171 |
+
task_prompt = '<REGION_TO_CATEGORY>'
|
172 |
+
results = run_example(task_prompt, image, text_input, model_id)
|
173 |
+
return results, None
|
174 |
+
elif task_prompt == 'Region to Description':
|
175 |
+
task_prompt = '<REGION_TO_DESCRIPTION>'
|
176 |
+
results = run_example(task_prompt, image, text_input, model_id)
|
177 |
+
return results, None
|
178 |
+
elif task_prompt == 'OCR':
|
179 |
+
task_prompt = '<OCR>'
|
180 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
181 |
+
return results, None
|
182 |
+
elif task_prompt == 'OCR with Region':
|
183 |
+
task_prompt = '<OCR_WITH_REGION>'
|
184 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
185 |
+
output_image = copy.deepcopy(image)
|
186 |
+
output_image = draw_ocr_bboxes(output_image, results['<OCR_WITH_REGION>'])
|
187 |
+
return results, output_image
|
188 |
+
else:
|
189 |
+
return "", None # Return empty string and None for unknown task prompts
|
190 |
+
|
191 |
+
css = """
|
192 |
+
#output {
|
193 |
+
height: 500px;
|
194 |
+
overflow: auto;
|
195 |
+
border: 1px solid #ccc;
|
196 |
+
}
|
197 |
+
"""
|
198 |
+
|
199 |
+
with gr.Blocks(css=css) as demo:
|
200 |
+
gr.Markdown(DESCRIPTION)
|
201 |
+
with gr.Tab(label="Florence-2 Image Captioning"):
|
202 |
+
with gr.Row():
|
203 |
+
with gr.Column():
|
204 |
+
input_img = gr.Image(label="Input Picture",height=400)
|
205 |
+
model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value='microsoft/Florence-2-base')
|
206 |
+
task_prompt = gr.Dropdown(choices=[
|
207 |
+
'Caption', 'Detailed Caption', 'More Detailed Caption', 'Object Detection',
|
208 |
+
'Dense Region Caption', 'Region Proposal', 'Caption to Phrase Grounding',
|
209 |
+
'Referring Expression Segmentation', 'Region to Segmentation',
|
210 |
+
'Open Vocabulary Detection', 'Region to Category', 'Region to Description',
|
211 |
+
'OCR', 'OCR with Region'
|
212 |
+
], label="Task Prompt", value= 'Caption')
|
213 |
+
text_input = gr.Textbox(label="Text Input (optional)")
|
214 |
+
submit_btn = gr.Button(value="Submit")
|
215 |
+
with gr.Column():
|
216 |
+
output_text = gr.Textbox(label="Output Text")
|
217 |
+
output_img = gr.Image(label="Output Image")
|
218 |
+
|
219 |
+
gr.Examples(
|
220 |
+
examples=[
|
221 |
+
["image1.jpg", 'Object Detection'],
|
222 |
+
["image2.jpg", 'OCR with Region']
|
223 |
+
],
|
224 |
+
inputs=[input_img, task_prompt],
|
225 |
+
outputs=[output_text, output_img],
|
226 |
+
fn=process_image,
|
227 |
+
cache_examples=True,
|
228 |
+
label='Try examples'
|
229 |
+
)
|
230 |
+
|
231 |
+
submit_btn.click(process_image, [input_img, task_prompt, text_input, model_selector], [output_text, output_img])
|
232 |
+
|
233 |
+
demo.launch(debug=True,server_name="0.0.0.0")
|
image1.jpg
ADDED
![]() |
image2.jpg
ADDED
![]() |
pre-requirements.txt
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
pip>=23.0.0
|
requirements.txt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
spaces
|
2 |
+
transformers
|
3 |
+
timm
|