app.py

import gradio as gr
import torch
from PIL import Image, ImageDraw
import numpy as np
import json
import base64
import io
import os
import secrets
from dotenv import load_dotenv
from megadetector.detection import run_detector

# Load environment variables for configuration
load_dotenv()

# Access token configuration
# You can set a fixed token in your Space's environment variables
# or generate a random one on startup (less secure)
API_TOKEN = os.getenv("API_TOKEN")
if not API_TOKEN:
    # Generate a random token if not provided - will change on restart!
    API_TOKEN = secrets.token_hex(16)
    print(f"Generated API token: {API_TOKEN}")
    print("IMPORTANT: This token will change if the space restarts!")
    print("Set a permanent token in the Space's environment variables.")

def validate_token(token):
    """Validate the provided access token"""
    return token == API_TOKEN

model = run_detector.load_detector('MDV5A')

# CVAT categories - customize based on your model's classes
CATEGORIES = [
    {"id": 1, "name": "animal"},
    {"id": 2, "name": "person"},
    {"id": 3, "name": "vehicle"},
    # Add all categories your model supports
]

def process_predictions(outputs, image, confidence_threshold=0.5):
    # Process the model outputs to match CVAT format
    results = []

    iw, ih = image.size

    for det in outputs['detections']:
        # Convert from [x, y, w, h] to [x1, y1, x2, y2]
        x, y, w, h = det['bbox']
        bbox = [x * iw, y * ih, (x + w) * iw, (y + h) * ih]
        score = det['conf']
        if score < confidence_threshold:
            continue
        # Convert to 0-indexed classes to match YOLOS
        label = int(det['category']) - 1
    
        category_id = int(label)
        category_name = CATEGORIES[category_id]["name"]
        
        result = {
            "confidence": float(score),
            "label": category_name,
            "points": [bbox[0], bbox[1], bbox[2], bbox[3]],
            "type": "rectangle"
        }
        results.append(result)

    return results

def predict(image_data, token=None):
    """Main prediction function for API endpoint
    
    Args:
        image_data: The image to be processed
        token: Access token for authentication
    """
    try:
        # Validate access token
        if token is None or not validate_token(token):
            return {"error": "Authentication failed. Invalid or missing access token."}
            
        # Handle various image input formats
        if isinstance(image_data, Image.Image):
            image = image_data
        elif isinstance(image_data, str) and image_data.startswith("data:image"):
            image_data = image_data.split(",")[1]
            image_bytes = base64.b64decode(image_data)
            image = Image.open(io.BytesIO(image_bytes))
        elif isinstance(image_data, np.ndarray):
            image = Image.fromarray(image_data)
        else:
            image = Image.open(image_data)
            
        # Process image with model
        outputs = model.generate_detections_one_image(image)
        
        # Process predictions
        results = process_predictions(outputs, image)
        
        # Return results in CVAT-compatible format
        return {"results": results}
    
    except Exception as e:
        return {"error": str(e)}

# Create Gradio interface for testing
def gradio_interface(image):
    # For the demo interface, we'll automatically pass the token
    results = predict(image, API_TOKEN)
    
    # Draw bounding boxes on image for visualization
    img_draw = image.copy()
    draw = ImageDraw.Draw(img_draw)
    
    for obj in results.get("results", []):
        box = obj["points"]
        draw.rectangle([box[0], box[1], box[2], box[3]], outline="red", width=3)
        draw.text((box[0], box[1]), f"{obj['label']} {obj['confidence']:.2f}", fill="red")
    
    return img_draw, json.dumps(results, indent=2)

# API endpoint for CVAT
app = gr.Interface(
    fn=predict,
    inputs=[
        gr.Image(type="filepath"),
        gr.Textbox(label="Access Token", type="password")
    ],
    outputs="json",
    title="Object Detection API for CVAT",
    description=f"Upload an image to get object detection predictions in CVAT-compatible format. Requires access token.",
    flagging_mode="never",
)

# UI for testing
demo = gr.Interface(
    fn=gradio_interface,
    inputs=gr.Image(type="pil"),
    outputs=[
        gr.Image(type="pil", label="Detection Result"),
        gr.JSON(label="JSON Output")
    ],
    title="Object Detection Demo",
    description="Test your object detection model with this interface",
    flagging_mode="never",
)

# Combine both interfaces
combined_demo = gr.TabbedInterface(
    [app, demo],
    ["API Endpoint", "Testing Interface"]
)

if __name__ == "__main__":
    combined_demo.launch()