Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,60 +1,91 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
import os
|
| 3 |
import json
|
| 4 |
-
from PIL import Image, ImageDraw, ImageFont
|
| 5 |
import google.generativeai as genai
|
| 6 |
|
| 7 |
-
# --- 1. SETUP
|
| 8 |
-
#
|
| 9 |
-
# (Note: For long-term safety, use Settings > Secrets, but this works for now!)
|
| 10 |
api_key = "AIzaSyABaidsygD73gbaSlMHlkrhhiDT8NDzrjE"
|
| 11 |
-
|
| 12 |
genai.configure(api_key=api_key)
|
| 13 |
|
| 14 |
-
|
| 15 |
-
generation_config = {
|
| 16 |
-
"temperature": 0.5,
|
| 17 |
-
"response_mime_type": "application/json"
|
| 18 |
-
}
|
| 19 |
-
|
| 20 |
-
system_instructions = """
|
| 21 |
-
Return bounding boxes as a JSON array.
|
| 22 |
-
Format: [{"box_2d": [ymin, xmin, ymax, xmax], "label": "string"}]
|
| 23 |
-
Coordinate scale: 0-1000.
|
| 24 |
-
Do not use markdown.
|
| 25 |
-
"""
|
| 26 |
-
|
| 27 |
-
model = genai.GenerativeModel(
|
| 28 |
-
model_name='gemini-1.5-flash',
|
| 29 |
-
system_instruction=system_instructions
|
| 30 |
-
)
|
| 31 |
-
|
| 32 |
-
# --- 3. HELPER FUNCTIONS ---
|
| 33 |
-
def parse_json(json_output):
|
| 34 |
-
"""Safely parse JSON even if the model adds text."""
|
| 35 |
-
try:
|
| 36 |
-
# Clean potential markdown
|
| 37 |
-
json_output = json_output.replace("```json", "").replace("```", "")
|
| 38 |
-
return json.loads(json_output)
|
| 39 |
-
except Exception as e:
|
| 40 |
-
print(f"JSON Parsing Error: {e}")
|
| 41 |
-
return []
|
| 42 |
|
|
|
|
| 43 |
def plot_bounding_boxes(im, boxes):
|
| 44 |
-
"""Draws boxes on the image."""
|
|
|
|
| 45 |
im = im.copy()
|
| 46 |
-
width, height = im.size
|
| 47 |
draw = ImageDraw.Draw(im)
|
|
|
|
| 48 |
|
| 49 |
-
#
|
| 50 |
-
colors = ['red', 'green', 'blue', 'yellow', 'cyan'
|
| 51 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
try:
|
| 53 |
-
#
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import os
|
| 3 |
import json
|
| 4 |
+
from PIL import Image, ImageDraw, ImageFont
|
| 5 |
import google.generativeai as genai
|
| 6 |
|
| 7 |
+
# --- 1. SETUP ---
|
| 8 |
+
# using the key you provided
|
|
|
|
| 9 |
api_key = "AIzaSyABaidsygD73gbaSlMHlkrhhiDT8NDzrjE"
|
|
|
|
| 10 |
genai.configure(api_key=api_key)
|
| 11 |
|
| 12 |
+
model = genai.GenerativeModel('gemini-1.5-flash')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
+
# --- 2. LOGIC ---
|
| 15 |
def plot_bounding_boxes(im, boxes):
|
| 16 |
+
"""Draws bounding boxes on the image."""
|
| 17 |
+
# Create a copy so we don't ruin the original
|
| 18 |
im = im.copy()
|
|
|
|
| 19 |
draw = ImageDraw.Draw(im)
|
| 20 |
+
width, height = im.size
|
| 21 |
|
| 22 |
+
# Simple list of colors
|
| 23 |
+
colors = ['red', 'green', 'blue', 'yellow', 'cyan']
|
| 24 |
|
| 25 |
+
for i, box in enumerate(boxes):
|
| 26 |
+
color = colors[i % len(colors)]
|
| 27 |
+
|
| 28 |
+
# Get the numbers (default to 0 if missing)
|
| 29 |
+
ymin, xmin, ymax, xmax = box.get("box_2d", [0,0,0,0])
|
| 30 |
+
label = box.get("label", "Object")
|
| 31 |
+
|
| 32 |
+
# Convert 0-1000 scale to pixels
|
| 33 |
+
left = int((xmin / 1000) * width)
|
| 34 |
+
top = int((ymin / 1000) * height)
|
| 35 |
+
right = int((xmax / 1000) * width)
|
| 36 |
+
bottom = int((ymax / 1000) * height)
|
| 37 |
+
|
| 38 |
+
# Draw the box
|
| 39 |
+
draw.rectangle([left, top, right, bottom], outline=color, width=4)
|
| 40 |
+
|
| 41 |
+
# Draw the text
|
| 42 |
+
# (We keep it simple to avoid font errors)
|
| 43 |
+
draw.text((left, top), label, fill=color)
|
| 44 |
+
|
| 45 |
+
return im
|
| 46 |
+
|
| 47 |
+
def detect(image):
|
| 48 |
+
if image is None:
|
| 49 |
+
return None, "Please upload an image."
|
| 50 |
+
|
| 51 |
+
# The prompt we send to Gemini
|
| 52 |
+
prompt = """
|
| 53 |
+
Detect objects in this image.
|
| 54 |
+
Return a JSON Array.
|
| 55 |
+
Format: [{"box_2d": [ymin, xmin, ymax, xmax], "label": "name"}]
|
| 56 |
+
"""
|
| 57 |
+
|
| 58 |
try:
|
| 59 |
+
# 1. Ask Gemini
|
| 60 |
+
response = model.generate_content([prompt, image])
|
| 61 |
+
text_data = response.text
|
| 62 |
+
|
| 63 |
+
# 2. Clean the text (remove markdown ```json ... ```)
|
| 64 |
+
text_data = text_data.replace("```json", "").replace("```", "")
|
| 65 |
+
|
| 66 |
+
# 3. Convert text to list
|
| 67 |
+
boxes = json.loads(text_data)
|
| 68 |
+
|
| 69 |
+
# 4. Draw boxes
|
| 70 |
+
result_image = plot_bounding_boxes(image, boxes)
|
| 71 |
+
|
| 72 |
+
return result_image, str(boxes)
|
| 73 |
+
|
| 74 |
+
except Exception as e:
|
| 75 |
+
return image, f"Error: {str(e)}"
|
| 76 |
+
|
| 77 |
+
# --- 3. UI ---
|
| 78 |
+
with gr.Blocks() as demo:
|
| 79 |
+
gr.Markdown("# 🔍 Gemini Object Detector")
|
| 80 |
+
|
| 81 |
+
with gr.Row():
|
| 82 |
+
inp = gr.Image(type="pil", label="Input Image")
|
| 83 |
+
out = gr.Image(type="pil", label="Result")
|
| 84 |
+
|
| 85 |
+
debug = gr.Textbox(label="Debug Information")
|
| 86 |
+
|
| 87 |
+
btn = gr.Button("Detect Objects")
|
| 88 |
+
btn.click(detect, inp, [out, debug])
|
| 89 |
+
|
| 90 |
+
if __name__ == "__main__":
|
| 91 |
+
demo.launch()
|