{"message":"Fashion Detection API v2","endpoints":{"detect":"POST /detect — multipart/form-data with 'file' or 'url' field","detect-url":"POST /detect-url — JSON body with {'url': '...'}","health":"GET /health — model and device status"},"improvements_in_v2":["SegFormer B5 model (upgraded from B2) for better segmentation","GPU support — automatically uses CUDA if available","Image resizing to 640px before inference","CLAHE contrast enhancement","Non-Maximum Suppression to remove duplicate detections","Lower accessory detection threshold (0.3 vs 0.5)","Model warm-up on startup to avoid slow first request","Crops returned at original image resolution","lifespan-based startup/shutdown (FastAPI best practice)"]}