The algorithm can distinguish PET, plastic, light_bulb, paper and cardboard garbages. It had 16 FPS in average on a Tesla P100-PCI-E-16GB
- Record the video
- Create a custom dataset by annotating 50 frames with CVAT
- Train yolov4 on the small dataset
- Generate pseudo labels with the trained model
- Import, and fix the generated iamges manually in CVAT
- Retrain the model on the larger dataset
- Process the video
- yolov4 Results: https://www.youtube.com/watch?v=iumbdbNbZsM
- EfficientDet D0
- yolov3