Spaces:
Sleeping
Sleeping
Delete app.py
Browse files
app.py
DELETED
@@ -1,130 +0,0 @@
|
|
1 |
-
import gradio as gr
|
2 |
-
import os
|
3 |
-
from io import BytesIO
|
4 |
-
from PIL import Image, ImageDraw, ImageFont
|
5 |
-
from PIL import ImageColor
|
6 |
-
import json
|
7 |
-
from google import genai
|
8 |
-
from google.genai import types
|
9 |
-
|
10 |
-
# Initialize Google Gemini client
|
11 |
-
client = genai.Client(api_key=os.environ['GOOGLE_API_KEY'])
|
12 |
-
model_name = "gemini-2.0-flash-exp"
|
13 |
-
|
14 |
-
|
15 |
-
#Skin issue detection system instructions
|
16 |
-
bounding_box_system_instructions = """
|
17 |
-
Return bounding boxes as a JSON array with labels specific to medical ultrasound imaging. Never return masks or code fencing. Limit to 25 objects per frame."""
|
18 |
-
|
19 |
-
# Additional colors f
|
20 |
-
|
21 |
-
# Additional colors for bounding box visualization
|
22 |
-
additional_colors = [colorname for (colorname, colorcode) in ImageColor.colormap.items()]
|
23 |
-
|
24 |
-
def parse_json(json_output):
|
25 |
-
"""
|
26 |
-
Parse JSON output from the Gemini model.
|
27 |
-
"""
|
28 |
-
try:
|
29 |
-
json_start = json_output.find("[")
|
30 |
-
json_end = json_output.rfind("]")
|
31 |
-
if json_start == -1 or json_end == -1:
|
32 |
-
raise ValueError("Bounding box JSON not found in response.")
|
33 |
-
return json_output[json_start:json_end + 1]
|
34 |
-
except Exception as e:
|
35 |
-
print(f"Error parsing JSON: {e}")
|
36 |
-
return None
|
37 |
-
|
38 |
-
def plot_bounding_boxes(im, bounding_boxes):
|
39 |
-
"""
|
40 |
-
Plots bounding boxes on an image with labels.
|
41 |
-
"""
|
42 |
-
im = im.copy()
|
43 |
-
draw = ImageDraw.Draw(im)
|
44 |
-
colors = ['red', 'green', 'blue', 'yellow', 'pink'] + additional_colors
|
45 |
-
|
46 |
-
try:
|
47 |
-
bounding_boxes_json = json.loads(bounding_boxes)
|
48 |
-
for i, bounding_box in enumerate(bounding_boxes_json):
|
49 |
-
color = colors[i % len(colors)]
|
50 |
-
x_min, y_min = bounding_box["x_min"], bounding_box["y_min"]
|
51 |
-
x_max, y_max = bounding_box["x_max"], bounding_box["y_max"]
|
52 |
-
draw.rectangle(((x_min, y_min), (x_max, y_max)), outline=color, width=4)
|
53 |
-
if "label" in bounding_box:
|
54 |
-
label = f"{bounding_box['label']} ({bounding_box['metadata']['severity']})"
|
55 |
-
draw.text((x_min + 5, y_min - 10), label, fill=color)
|
56 |
-
except Exception as e:
|
57 |
-
print(f"Error drawing bounding boxes: {e}")
|
58 |
-
|
59 |
-
return im
|
60 |
-
|
61 |
-
def predict_bounding_boxes(image, prompt):
|
62 |
-
"""
|
63 |
-
Process the image and prompt through Gemini and draw bounding boxes.
|
64 |
-
"""
|
65 |
-
try:
|
66 |
-
image = image.resize((1024, int(1024 * image.height / image.width)))
|
67 |
-
buffered = BytesIO()
|
68 |
-
image.save(buffered, format="JPEG")
|
69 |
-
image_bytes = buffered.getvalue()
|
70 |
-
|
71 |
-
response = client.models.generate_content(
|
72 |
-
model=model_name,
|
73 |
-
contents=[prompt, image],
|
74 |
-
config=types.GenerateContentConfig(
|
75 |
-
system_instruction=bounding_box_system_instructions,
|
76 |
-
temperature=0.5
|
77 |
-
)
|
78 |
-
)
|
79 |
-
|
80 |
-
bounding_boxes = parse_json(response.text)
|
81 |
-
if not bounding_boxes:
|
82 |
-
raise ValueError("No bounding boxes returned.")
|
83 |
-
return plot_bounding_boxes(image, bounding_boxes)
|
84 |
-
except Exception as e:
|
85 |
-
print(f"Error: {e}")
|
86 |
-
return image
|
87 |
-
|
88 |
-
def gradio_interface():
|
89 |
-
"""
|
90 |
-
Gradio app interface for skin issue detection.
|
91 |
-
"""
|
92 |
-
examples = [
|
93 |
-
["skin_example1.jpg", "Detect and label skin abnormalities."],
|
94 |
-
["skin_example2.jpg", "Find acne lesions and classify severity."],
|
95 |
-
["skin_example3.jpg", "Identify suspicious moles and discolorations."],
|
96 |
-
]
|
97 |
-
|
98 |
-
with gr.Blocks() as demo:
|
99 |
-
gr.Markdown("# Skin Issue Detection with Gemini")
|
100 |
-
|
101 |
-
with gr.Row():
|
102 |
-
with gr.Column():
|
103 |
-
input_image = gr.Image(type="pil", label="Input Image")
|
104 |
-
input_prompt = gr.Textbox(
|
105 |
-
lines=2,
|
106 |
-
label="Input Prompt",
|
107 |
-
placeholder="Describe what to detect (e.g., 'Identify acne lesions').",
|
108 |
-
value="Identify and label skin abnormalities."
|
109 |
-
)
|
110 |
-
submit_btn = gr.Button("Generate")
|
111 |
-
|
112 |
-
with gr.Column():
|
113 |
-
output_image = gr.Image(type="pil", label="Output Image")
|
114 |
-
|
115 |
-
gr.Examples(
|
116 |
-
examples=examples,
|
117 |
-
inputs=[input_image, input_prompt]
|
118 |
-
)
|
119 |
-
|
120 |
-
submit_btn.click(
|
121 |
-
predict_bounding_boxes,
|
122 |
-
inputs=[input_image, input_prompt],
|
123 |
-
outputs=[output_image]
|
124 |
-
)
|
125 |
-
|
126 |
-
return demo
|
127 |
-
|
128 |
-
if __name__ == "__main__":
|
129 |
-
app = gradio_interface()
|
130 |
-
app.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|