Spaces:
Sleeping
Sleeping
Commit
•
d502400
1
Parent(s):
b9bfc9d
Update app.py
Browse files
app.py
CHANGED
@@ -16,9 +16,19 @@ import numpy as np
|
|
16 |
import subprocess
|
17 |
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
|
18 |
|
19 |
-
|
20 |
-
|
21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
|
23 |
|
24 |
DESCRIPTION = "# [Florence-2 Demo](https://huggingface.co/microsoft/Florence-2-large)"
|
@@ -32,8 +42,10 @@ def fig_to_pil(fig):
|
|
32 |
buf.seek(0)
|
33 |
return Image.open(buf)
|
34 |
|
35 |
-
|
36 |
-
def run_example(task_prompt, image, text_input=None):
|
|
|
|
|
37 |
if text_input is None:
|
38 |
prompt = task_prompt
|
39 |
else:
|
@@ -109,73 +121,73 @@ def draw_ocr_bboxes(image, prediction):
|
|
109 |
fill=color)
|
110 |
return image
|
111 |
|
112 |
-
def process_image(image, task_prompt, text_input=None):
|
113 |
image = Image.fromarray(image) # Convert NumPy array to PIL Image
|
114 |
if task_prompt == 'Caption':
|
115 |
task_prompt = '<CAPTION>'
|
116 |
-
|
117 |
-
return
|
118 |
elif task_prompt == 'Detailed Caption':
|
119 |
task_prompt = '<DETAILED_CAPTION>'
|
120 |
-
|
121 |
-
return
|
122 |
elif task_prompt == 'More Detailed Caption':
|
123 |
task_prompt = '<MORE_DETAILED_CAPTION>'
|
124 |
-
|
125 |
-
return
|
126 |
elif task_prompt == 'Object Detection':
|
127 |
task_prompt = '<OD>'
|
128 |
-
results = run_example(task_prompt, image)
|
129 |
fig = plot_bbox(image, results['<OD>'])
|
130 |
return results, fig_to_pil(fig)
|
131 |
elif task_prompt == 'Dense Region Caption':
|
132 |
task_prompt = '<DENSE_REGION_CAPTION>'
|
133 |
-
results = run_example(task_prompt, image)
|
134 |
fig = plot_bbox(image, results['<DENSE_REGION_CAPTION>'])
|
135 |
return results, fig_to_pil(fig)
|
136 |
elif task_prompt == 'Region Proposal':
|
137 |
task_prompt = '<REGION_PROPOSAL>'
|
138 |
-
results = run_example(task_prompt, image)
|
139 |
fig = plot_bbox(image, results['<REGION_PROPOSAL>'])
|
140 |
return results, fig_to_pil(fig)
|
141 |
elif task_prompt == 'Caption to Phrase Grounding':
|
142 |
task_prompt = '<CAPTION_TO_PHRASE_GROUNDING>'
|
143 |
-
results = run_example(task_prompt, image, text_input)
|
144 |
fig = plot_bbox(image, results['<CAPTION_TO_PHRASE_GROUNDING>'])
|
145 |
return results, fig_to_pil(fig)
|
146 |
elif task_prompt == 'Referring Expression Segmentation':
|
147 |
task_prompt = '<REFERRING_EXPRESSION_SEGMENTATION>'
|
148 |
-
results = run_example(task_prompt, image, text_input)
|
149 |
output_image = copy.deepcopy(image)
|
150 |
output_image = draw_polygons(output_image, results['<REFERRING_EXPRESSION_SEGMENTATION>'], fill_mask=True)
|
151 |
return results, output_image
|
152 |
elif task_prompt == 'Region to Segmentation':
|
153 |
task_prompt = '<REGION_TO_SEGMENTATION>'
|
154 |
-
results = run_example(task_prompt, image, text_input)
|
155 |
output_image = copy.deepcopy(image)
|
156 |
output_image = draw_polygons(output_image, results['<REGION_TO_SEGMENTATION>'], fill_mask=True)
|
157 |
return results, output_image
|
158 |
elif task_prompt == 'Open Vocabulary Detection':
|
159 |
task_prompt = '<OPEN_VOCABULARY_DETECTION>'
|
160 |
-
results = run_example(task_prompt, image, text_input)
|
161 |
bbox_results = convert_to_od_format(results['<OPEN_VOCABULARY_DETECTION>'])
|
162 |
fig = plot_bbox(image, bbox_results)
|
163 |
return results, fig_to_pil(fig)
|
164 |
elif task_prompt == 'Region to Category':
|
165 |
task_prompt = '<REGION_TO_CATEGORY>'
|
166 |
-
results = run_example(task_prompt, image, text_input)
|
167 |
return results, None
|
168 |
elif task_prompt == 'Region to Description':
|
169 |
task_prompt = '<REGION_TO_DESCRIPTION>'
|
170 |
-
results = run_example(task_prompt, image, text_input)
|
171 |
return results, None
|
172 |
elif task_prompt == 'OCR':
|
173 |
task_prompt = '<OCR>'
|
174 |
-
|
175 |
-
return
|
176 |
elif task_prompt == 'OCR with Region':
|
177 |
task_prompt = '<OCR_WITH_REGION>'
|
178 |
-
results = run_example(task_prompt, image)
|
179 |
output_image = copy.deepcopy(image)
|
180 |
output_image = draw_ocr_bboxes(output_image, results['<OCR_WITH_REGION>'])
|
181 |
return results, output_image
|
@@ -196,6 +208,7 @@ with gr.Blocks(css=css) as demo:
|
|
196 |
with gr.Row():
|
197 |
with gr.Column():
|
198 |
input_img = gr.Image(label="Input Picture")
|
|
|
199 |
task_prompt = gr.Dropdown(choices=[
|
200 |
'Caption', 'Detailed Caption', 'More Detailed Caption', 'Object Detection',
|
201 |
'Dense Region Caption', 'Region Proposal', 'Caption to Phrase Grounding',
|
@@ -221,6 +234,6 @@ with gr.Blocks(css=css) as demo:
|
|
221 |
label='Try examples'
|
222 |
)
|
223 |
|
224 |
-
submit_btn.click(process_image, [input_img, task_prompt, text_input], [output_text, output_img])
|
225 |
|
226 |
demo.launch(debug=True)
|
|
|
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-large-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large-ft', trust_remote_code=True).to("cuda").eval(),
|
21 |
+
'microsoft/Florence-2-large': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True).to("cuda").eval(),
|
22 |
+
'microsoft/Florence-2-base-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base-ft', trust_remote_code=True).to("cuda").eval(),
|
23 |
+
'microsoft/Florence-2-base': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True).to("cuda").eval(),
|
24 |
+
}
|
25 |
+
|
26 |
+
processors = {
|
27 |
+
'microsoft/Florence-2-large-ft': AutoProcessor.from_pretrained('microsoft/Florence-2-large-ft', trust_remote_code=True),
|
28 |
+
'microsoft/Florence-2-large': AutoProcessor.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True),
|
29 |
+
'microsoft/Florence-2-base-ft': AutoProcessor.from_pretrained('microsoft/Florence-2-base-ft', trust_remote_code=True),
|
30 |
+
'microsoft/Florence-2-base': AutoProcessor.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True),
|
31 |
+
}
|
32 |
|
33 |
|
34 |
DESCRIPTION = "# [Florence-2 Demo](https://huggingface.co/microsoft/Florence-2-large)"
|
|
|
42 |
buf.seek(0)
|
43 |
return Image.open(buf)
|
44 |
|
45 |
+
|
46 |
+
def run_example(task_prompt, image, text_input=None, model_id='microsoft/Florence-2-large'):
|
47 |
+
model = models[model_id]
|
48 |
+
processor = processors[model_id]
|
49 |
if text_input is None:
|
50 |
prompt = task_prompt
|
51 |
else:
|
|
|
121 |
fill=color)
|
122 |
return image
|
123 |
|
124 |
+
def process_image(image, task_prompt, text_input=None, model_id='microsoft/Florence-2-large'):
|
125 |
image = Image.fromarray(image) # Convert NumPy array to PIL Image
|
126 |
if task_prompt == 'Caption':
|
127 |
task_prompt = '<CAPTION>'
|
128 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
129 |
+
return results, None
|
130 |
elif task_prompt == 'Detailed Caption':
|
131 |
task_prompt = '<DETAILED_CAPTION>'
|
132 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
133 |
+
return results, None
|
134 |
elif task_prompt == 'More Detailed Caption':
|
135 |
task_prompt = '<MORE_DETAILED_CAPTION>'
|
136 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
137 |
+
return results, None
|
138 |
elif task_prompt == 'Object Detection':
|
139 |
task_prompt = '<OD>'
|
140 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
141 |
fig = plot_bbox(image, results['<OD>'])
|
142 |
return results, fig_to_pil(fig)
|
143 |
elif task_prompt == 'Dense Region Caption':
|
144 |
task_prompt = '<DENSE_REGION_CAPTION>'
|
145 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
146 |
fig = plot_bbox(image, results['<DENSE_REGION_CAPTION>'])
|
147 |
return results, fig_to_pil(fig)
|
148 |
elif task_prompt == 'Region Proposal':
|
149 |
task_prompt = '<REGION_PROPOSAL>'
|
150 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
151 |
fig = plot_bbox(image, results['<REGION_PROPOSAL>'])
|
152 |
return results, fig_to_pil(fig)
|
153 |
elif task_prompt == 'Caption to Phrase Grounding':
|
154 |
task_prompt = '<CAPTION_TO_PHRASE_GROUNDING>'
|
155 |
+
results = run_example(task_prompt, image, text_input, model_id)
|
156 |
fig = plot_bbox(image, results['<CAPTION_TO_PHRASE_GROUNDING>'])
|
157 |
return results, fig_to_pil(fig)
|
158 |
elif task_prompt == 'Referring Expression Segmentation':
|
159 |
task_prompt = '<REFERRING_EXPRESSION_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['<REFERRING_EXPRESSION_SEGMENTATION>'], fill_mask=True)
|
163 |
return results, output_image
|
164 |
elif task_prompt == 'Region to Segmentation':
|
165 |
task_prompt = '<REGION_TO_SEGMENTATION>'
|
166 |
+
results = run_example(task_prompt, image, text_input, model_id)
|
167 |
output_image = copy.deepcopy(image)
|
168 |
output_image = draw_polygons(output_image, results['<REGION_TO_SEGMENTATION>'], fill_mask=True)
|
169 |
return results, output_image
|
170 |
elif task_prompt == 'Open Vocabulary Detection':
|
171 |
task_prompt = '<OPEN_VOCABULARY_DETECTION>'
|
172 |
+
results = run_example(task_prompt, image, text_input, model_id)
|
173 |
bbox_results = convert_to_od_format(results['<OPEN_VOCABULARY_DETECTION>'])
|
174 |
fig = plot_bbox(image, bbox_results)
|
175 |
return results, fig_to_pil(fig)
|
176 |
elif task_prompt == 'Region to Category':
|
177 |
task_prompt = '<REGION_TO_CATEGORY>'
|
178 |
+
results = run_example(task_prompt, image, text_input, model_id)
|
179 |
return results, None
|
180 |
elif task_prompt == 'Region to Description':
|
181 |
task_prompt = '<REGION_TO_DESCRIPTION>'
|
182 |
+
results = run_example(task_prompt, image, text_input, model_id)
|
183 |
return results, None
|
184 |
elif task_prompt == 'OCR':
|
185 |
task_prompt = '<OCR>'
|
186 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
187 |
+
return results, None
|
188 |
elif task_prompt == 'OCR with Region':
|
189 |
task_prompt = '<OCR_WITH_REGION>'
|
190 |
+
results = run_example(task_prompt, image, model_id=model_id)
|
191 |
output_image = copy.deepcopy(image)
|
192 |
output_image = draw_ocr_bboxes(output_image, results['<OCR_WITH_REGION>'])
|
193 |
return results, output_image
|
|
|
208 |
with gr.Row():
|
209 |
with gr.Column():
|
210 |
input_img = gr.Image(label="Input Picture")
|
211 |
+
model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value='microsoft/Florence-2-large')
|
212 |
task_prompt = gr.Dropdown(choices=[
|
213 |
'Caption', 'Detailed Caption', 'More Detailed Caption', 'Object Detection',
|
214 |
'Dense Region Caption', 'Region Proposal', 'Caption to Phrase Grounding',
|
|
|
234 |
label='Try examples'
|
235 |
)
|
236 |
|
237 |
+
submit_btn.click(process_image, [input_img, task_prompt, text_input, model_selector], [output_text, output_img])
|
238 |
|
239 |
demo.launch(debug=True)
|