ssd_tensorflow_traffic_sign_detection

Why we included this project

Developers who want to study how the Single Shot MultiBox Detector is assembled in TensorFlow will find this a useful reference. The code is split into a few straightforward scripts that take you from data preparation through training to inference, and the project targets one narrow job: finding and classifying traffic signs in images and video. The author is upfront about the limits: it only recognizes stop and pedestrian-crossing signs, it was trained on the LISA US traffic sign dataset, and the model overfits, so treat it as a learning base or a starting point for transfer learning rather than production software. The reported 40-45 fps on a GTX 1080 also gives a rough idea of the real-time headroom available once the NMS step is optimized.