- Download YoloV5 weights folder from [here](https://gitlab.tugraz.at/2EB2687AA157E3CC/anpr/-/tree/weights/OpenImages) and put in ```weights/plate_detector```
- Download ViTSTR ```vitstr_small_patch16_224-Seed42``` and ```vitstr_tiny_patch16_224-Seed42``` weights folders from [here](https://gitlab.tugraz.at/2EB2687AA157E3CC/anpr/-/tree/weights/) and put in ```weights/plate_recognizer```
- Use this ViTSTR [weight](https://gitlab.tugraz.at/2EB2687AA157E3CC/anpr/-/blob/weights/real_synth/vitstr_tiny_patch16_224-Seed42/best_accuracy.pth)
- Use this YOLOv5 [weight](https://gitlab.tugraz.at/2EB2687AA157E3CC/anpr/-/blob/weights/OpenImages/yolov5n6_best.pt)
## YoloV5 Training log
- We tried training using OpenImages (all models) and CCPD (only YoloV5n6 due to large training time, big dataset). Check logs [here](https://wandb.ai/ivu_practical/projects)
