UPDATE

1. Add ONNX Model Inference;
2. Add Gradio Demo.

OPTIMIZE

1. Training license plate segmentation algorithm yolov5n-seg based on ChineseLicensePlate;
2. Using CRNN_Tiny for license plate recognition, lighter and more powerful.

Detect+Seg+Recog

Implementing license plate detection, segmentation, and recognition functions based on YOLOv5-v7.0 and CRNN-CTC

Training with AMP

• Support for Automatic Mixed Precision (AMP) training.

Training with Multi-GPUs

• Support for distributed training with multi-GPUs;
• Fix CCPDEvaluator: reset() required before each round of evaluation.

Init Version

• Reconstruct the 872699467/CCPD_CNN implementation to adapt to interfaces after Pytorch v1.0.0;
• Train and evaluate based on the detectRecog/CCPD dataset.