MazenBahgat commited on
Commit
09cb914
·
verified ·
1 Parent(s): cf21691

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +85 -85
app.py CHANGED
@@ -1,86 +1,86 @@
1
- import gradio as gr
2
- import os
3
- import json
4
- import google.generativeai as genai
5
- from PIL import Image, ImageDraw, ImageFont, ImageColor
6
- from dotenv import load_dotenv
7
-
8
- load_dotenv()
9
- api_key = os.getenv("Gemini_API_Key")
10
- genai.configure(api_key=api_key)
11
-
12
- SYSTEM_INSTRUCTION = """
13
- Return bounding boxes as a JSON array with labels. Never return masks or code fencing. Limit to 25 objects.
14
- If an object is present multiple times, name them according to their unique characteristic.
15
- Use the format: [{"box_2d": [ymin, xmin, ymax, xmax], "label": "object_name"}]
16
- """
17
-
18
- model = genai.GenerativeModel(
19
- model_name='gemini-2.5-flash',
20
- system_instruction=SYSTEM_INSTRUCTION
21
- )
22
-
23
- def parse_response(text):
24
- """Cleans markdown formatting if the model includes it."""
25
- clean = text.replace("json", "").replace("", "").strip()
26
- return clean
27
-
28
- def draw_boxes(image, boxes_json):
29
- """Draws boxes on the image using normalized coordinates (0-1000)."""
30
- img = image.copy()
31
- draw = ImageDraw.Draw(img)
32
- width, height = img.size
33
- colors = ["red", "green", "blue", "yellow", "magenta", "cyan", "orange"]
34
-
35
- try:
36
- data = json.loads(boxes_json)
37
- for i, item in enumerate(data):
38
- color = colors[i % len(colors)]
39
- # Gemini order: [ymin, xmin, ymax, xmax]
40
- ymin, xmin, ymax, xmax = item["box_2d"]
41
-
42
- left = (xmin / 1000) * width
43
- top = (ymin / 1000) * height
44
- right = (xmax / 1000) * width
45
- bottom = (ymax / 1000) * height
46
-
47
- draw.rectangle([left, top, right, bottom], outline=color, width=5)
48
- draw.text((left + 5, top + 5), item.get("label", "Object"), fill=color)
49
- except Exception as e:
50
- print(f"Error drawing boxes: {e}")
51
-
52
- return img
53
-
54
- def process_image(input_img, user_instruction):
55
- if input_img is None:
56
- return None
57
-
58
- if not user_instruction.strip():
59
- user_instruction = "Identify and label the objects in the image."
60
-
61
- # Combine user instruction with JSON formatting requirement
62
- full_prompt = f"{user_instruction} Return ONLY the JSON array of bounding boxes."
63
-
64
- response = model.generate_content([full_prompt, input_img])
65
- json_data = parse_response(response.text)
66
-
67
- output_img = draw_boxes(input_img, json_data)
68
- return output_img
69
-
70
- demo = gr.Interface(
71
- fn=process_image,
72
- inputs=[
73
- gr.Image(type="pil", label="Upload Image"),
74
- gr.Textbox(
75
- label="What should I find?",
76
- placeholder="e.g., 'Find all the red apples' or 'Detect the cars and pedestrians'",
77
- lines=2
78
- )
79
- ],
80
- outputs=gr.Image(type="pil", label="Image with Bounding Boxes"),
81
- title="Dynamic Gemini Object Detector",
82
- description="Upload an image and give specific instructions on what to detect."
83
- )
84
-
85
- if __name__ == "__main__":
86
  demo.launch()
 
1
+ import gradio as gr
2
+ import os
3
+ import json
4
+ import google.generativeai as genai
5
+ from PIL import Image, ImageDraw, ImageFont, ImageColor
6
+ from dotenv import load_dotenv
7
+
8
+ load_dotenv()
9
+ api_key = os.getenv("Gemini_API_Key")
10
+ genai.configure(api_key=api_key)
11
+
12
+ SYSTEM_INSTRUCTION = """
13
+ Return bounding boxes as a JSON array with labels. Never return masks or code fencing. Limit to 25 objects.
14
+ If an object is present multiple times, name them according to their unique characteristic.
15
+ Use the format: [{"box_2d": [ymin, xmin, ymax, xmax], "label": "object_name"}]
16
+ """
17
+
18
+ model = genai.GenerativeModel(
19
+ model_name='gemini-2.5-flash',
20
+ system_instruction=SYSTEM_INSTRUCTION
21
+ )
22
+
23
+ def parse_response(text):
24
+ """Cleans markdown formatting if the model includes it."""
25
+ clean = text.replace("json", "").replace("", "").strip()
26
+ return clean
27
+
28
+ def draw_boxes(image, boxes_json):
29
+ """Draws boxes on the image using normalized coordinates (0-1000)."""
30
+ img = image.copy()
31
+ draw = ImageDraw.Draw(img)
32
+ width, height = img.size
33
+ colors = ["red", "green", "blue", "yellow", "magenta", "cyan", "orange"]
34
+
35
+ try:
36
+ data = json.loads(boxes_json)
37
+ for i, item in enumerate(data):
38
+ color = colors[i % len(colors)]
39
+ # Gemini order: [ymin, xmin, ymax, xmax]
40
+ ymin, xmin, ymax, xmax = item["box_2d"]
41
+
42
+ left = (xmin / 1000) * width
43
+ top = (ymin / 1000) * height
44
+ right = (xmax / 1000) * width
45
+ bottom = (ymax / 1000) * height
46
+
47
+ draw.rectangle([left, top, right, bottom], outline=color, width=5)
48
+ draw.text((left + 5, top + 5), item.get("label", "Object"), fill=color)
49
+ except Exception as e:
50
+ print(f"Error drawing boxes: {e}")
51
+
52
+ return img
53
+
54
+ def process_image(input_img, user_instruction):
55
+ if input_img is None:
56
+ return None
57
+
58
+ if not user_instruction.strip():
59
+ user_instruction = "Identify and label the objects in the image."
60
+
61
+ # Combine user instruction with JSON formatting requirement
62
+ full_prompt = f"{user_instruction} Return ONLY the JSON array of bounding boxes."
63
+
64
+ response = model.generate_content([full_prompt, input_img])
65
+ json_data = parse_response(response.text)
66
+
67
+ output_img = draw_boxes(input_img, json_data)
68
+ return output_img
69
+
70
+ demo = gr.Interface(
71
+ fn=process_image,
72
+ inputs=[
73
+ gr.Image(type="pil", label="Upload Image"),
74
+ gr.Textbox(
75
+ label="What should I find?",
76
+ placeholder="e.g., 'Find all the red apples' or 'Detect the cars and pedestrians'",
77
+ lines=2
78
+ )
79
+ ],
80
+ outputs=gr.Image(type="pil", label="Image with Bounding Boxes"),
81
+ title="Your own Object Detector",
82
+ description="Upload an image and give specific instructions on what to detect."
83
+ )
84
+
85
+ if __name__ == "__main__":
86
  demo.launch()