Spaces:
Running
on
Zero
Running
on
Zero
Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,118 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import gradio as gr
|
3 |
+
from transformers import AutoModelForCausalLM, AutoProcessor
|
4 |
+
from PIL import ImageDraw
|
5 |
+
|
6 |
+
|
7 |
+
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
8 |
+
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
9 |
+
|
10 |
+
models = {
|
11 |
+
"AskUI/PTA-1": AutoModelForCausalLM.from_pretrained("AskUI/PTA-1", trust_remote_code=True),
|
12 |
+
}
|
13 |
+
|
14 |
+
processors = {
|
15 |
+
"AskUI/PTA-1": AutoProcessor.from_pretrained("AskUI/PTA-1", trust_remote_code=True)
|
16 |
+
}
|
17 |
+
|
18 |
+
|
19 |
+
def draw_bounding_boxes(image, bounding_boxes, outline_color="red", line_width=3):
|
20 |
+
draw = ImageDraw.Draw(image)
|
21 |
+
for box in bounding_boxes:
|
22 |
+
xmin, ymin, xmax, ymax = box
|
23 |
+
draw.rectangle([xmin, ymin, xmax, ymax], outline=outline_color, width=line_width)
|
24 |
+
return image
|
25 |
+
|
26 |
+
|
27 |
+
def florence_output_to_box(output):
|
28 |
+
try:
|
29 |
+
if "polygons" in output and len(output["polygons"]) > 0:
|
30 |
+
polygons = output["polygons"]
|
31 |
+
target_polygon = polygons[0][0]
|
32 |
+
target_polygon = [int(el) for el in target_polygon]
|
33 |
+
return [
|
34 |
+
target_polygon[0],
|
35 |
+
target_polygon[1],
|
36 |
+
target_polygon[4],
|
37 |
+
target_polygon[5],
|
38 |
+
]
|
39 |
+
if "bboxes" in output and len(output["bboxes"]) > 0:
|
40 |
+
bboxes = output["bboxes"]
|
41 |
+
target_bbox = bboxes[0]
|
42 |
+
target_bbox = [int(el) for el in target_bbox]
|
43 |
+
return target_bbox
|
44 |
+
except Exception as e:
|
45 |
+
print(f"Error: {e}")
|
46 |
+
return None
|
47 |
+
|
48 |
+
|
49 |
+
def run_example(image, text_input, model_id="AskUI/PTA-1"):
|
50 |
+
model = models[model_id].to(device, torch_dtype)
|
51 |
+
processor = processors[model_id]
|
52 |
+
task_prompt = "<OPEN_VOCABULARY_DETECTION>"
|
53 |
+
prompt = task_prompt + text_input
|
54 |
+
|
55 |
+
image = image.convert("RGB")
|
56 |
+
|
57 |
+
inputs = processor(text=prompt, images=image, return_tensors="pt").to(device, torch_dtype)
|
58 |
+
|
59 |
+
generated_ids = model.generate(
|
60 |
+
input_ids=inputs["input_ids"],
|
61 |
+
pixel_values=inputs["pixel_values"],
|
62 |
+
max_new_tokens=1024,
|
63 |
+
do_sample=False,
|
64 |
+
num_beams=3,
|
65 |
+
)
|
66 |
+
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
|
67 |
+
parsed_answer = processor.post_process_generation(generated_text, task="<OPEN_VOCABULARY_DETECTION>", image_size=(image.width, image.height))
|
68 |
+
target_box = florence_output_to_box(parsed_answer["<OPEN_VOCABULARY_DETECTION>"])
|
69 |
+
return target_box, draw_bounding_boxes(image, [target_box])
|
70 |
+
|
71 |
+
|
72 |
+
css = """
|
73 |
+
#output {
|
74 |
+
height: 500px;
|
75 |
+
overflow: auto;
|
76 |
+
border: 1px solid #ccc;
|
77 |
+
}
|
78 |
+
"""
|
79 |
+
with gr.Blocks(css=css) as demo:
|
80 |
+
gr.Markdown(
|
81 |
+
"""
|
82 |
+
<div style="display: flex; justify-content: space-between; align-items: center; background-color: #baff49; padding: 10px;">
|
83 |
+
<h1 style="margin: 0; color: #101828";>PTA-1: Controlling Computers with Small Models</h1>
|
84 |
+
<img src="https://cdn.prod.website-files.com/6627a15f6d261b8bf852c0a1/670529b583d3638f72db5614_askui-logo-primary-filled.svg" alt="Logo" style="height: 50px;">
|
85 |
+
</div>
|
86 |
+
""")
|
87 |
+
gr.Markdown("Check out the model [AskUI/PTA-1](https://huggingface.co/AskUI/PTA-1).")
|
88 |
+
with gr.Row():
|
89 |
+
with gr.Column():
|
90 |
+
input_img = gr.Image(label="Input Image", type="pil")
|
91 |
+
model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value="AskUI/PTA-1")
|
92 |
+
text_input = gr.Textbox(label="User Prompt")
|
93 |
+
submit_btn = gr.Button(value="Submit")
|
94 |
+
with gr.Column():
|
95 |
+
model_output_text = gr.Textbox(label="Model Output Text")
|
96 |
+
annotated_image = gr.Image(label="Annotated Image")
|
97 |
+
|
98 |
+
gr.Examples(
|
99 |
+
examples=[
|
100 |
+
["assets/sample.png", "search box"],
|
101 |
+
["assets/sample.png", "Query Service"],
|
102 |
+
["assets/ipad.png", "App Store icon"],
|
103 |
+
["assets/ipad.png", 'colorful icon with letter "S"'],
|
104 |
+
["assets/phone.jpg", "password field"],
|
105 |
+
["assets/phone.jpg", "back arrow icon"],
|
106 |
+
["assets/windows.jpg", "icon with letter S"],
|
107 |
+
["assets/windows.jpg", "Settings"],
|
108 |
+
],
|
109 |
+
inputs=[input_img, text_input],
|
110 |
+
outputs=[model_output_text, annotated_image],
|
111 |
+
fn=run_example,
|
112 |
+
cache_examples=False,
|
113 |
+
label="Try examples"
|
114 |
+
)
|
115 |
+
|
116 |
+
submit_btn.click(run_example, [input_img, text_input, model_selector], [model_output_text, annotated_image])
|
117 |
+
|
118 |
+
demo.launch(debug=False)
|