Pothole Detection (YOLOX-s, fine-tuned)

Fine-tuned YOLOX-s (Apache-2.0, Megvii) checkpoint for single-class pothole detection, trained on the jaygala24/pothole-detection dataset (MIT licensed).

pothole_yolox_s.pth -- fine-tuned PyTorch checkpoint
pothole_yolox_s.onnx -- static-shape (640x640) ONNX export, raw decoupled-head output (decoding happens client-side; see onnx_utils.py in the source repo)

Measured on the 186-image validation split: mAP@0.5 = 72.1%, mAP@0.5:0.95 = 34.4%.

Source: models/pothole-detection in AMD's physical_ai_sdk.

Downloads last month: -; Downloads are not tracked for this model. How to track