backend_app.py

import os
import json
import tempfile
from fastapi import FastAPI, UploadFile, File
from fastapi.responses import FileResponse
from ultralytics import YOLO
import supervision as sv

# --- MODEL AND APP INITIALIZATION ---
# The model is loaded only ONCE when the server starts, making it fast.
MODEL_PATH = "best.pt"
model = YOLO(MODEL_PATH)

app = FastAPI(title="YOLOv8 Tracking API")

# --- YOUR EXISTING PROCESSING LOGIC (UNCHANGED) ---
# We just put your code inside this function.
def process_video_logic(input_path, output_path, json_path):
    tracker = sv.ByteTrack()
    box_annotator = sv.BoxAnnotator(thickness=5)
    label_annotator = sv.LabelAnnotator(text_position=sv.Position.TOP_CENTER, text_scale=1, text_thickness=1)
    frame_generator = sv.get_video_frames_generator(source_path=input_path)
    video_info = sv.VideoInfo.from_video_path(input_path)
    results_list = []

    with sv.VideoSink(target_path=output_path, video_info=video_info) as sink:
        for frame_number, frame in enumerate(frame_generator):
            results = model(frame, verbose=False)[0]
            detections = sv.Detections.from_ultralytics(results)
            tracked_detections = tracker.update_with_detections(detections=detections)
            labels = [f"ID: {det[4]} {model.model.names[int(det[3])]}" for det in tracked_detections]
            annotated_frame = box_annotator.annotate(scene=frame.copy(), detections=tracked_detections)
            annotated_frame = label_annotator.annotate(scene=annotated_frame, detections=tracked_detections, labels=labels)
            
            for det in tracked_detections:
                bbox, conf, class_id, tracker_id = det[0], det[2], int(det[3]), det[4]
                results_list.append({
                    "frame_number": frame_number,
                    "track_id": int(tracker_id),
                    "class": model.model.names[class_id],
                    "confidence": float(conf),
                    "bounding_box": [int(coord) for coord in bbox]
                })
            sink.write_frame(frame=annotated_frame)

    with open(json_path, 'w') as f:
        json.dump(results_list, f, indent=4)

# --- API ENDPOINT ---
@app.post("/track/")
async def track_video_endpoint(video: UploadFile = File(...)):
    # Use a temporary directory to handle file operations safely
    with tempfile.TemporaryDirectory() as temp_dir:
        input_path = os.path.join(temp_dir, video.filename)
        output_video_path = os.path.join(temp_dir, "output.mp4")
        output_json_path = os.path.join(temp_dir, "results.json")

        # Save the uploaded video file
        with open(input_path, "wb") as buffer:
            buffer.write(await video.read())

        # Run your existing processing logic
        process_video_logic(input_path, output_video_path, output_json_path)

        # Return the processed video as a downloadable file
        return FileResponse(output_video_path, media_type="video/mp4", filename="output.mp4")