① real pruning project support, quantify and knowledge distilled YOLOV3
[Features]
1, after providing a plurality of pre-set main target detection data files and training method.
2, provided include pruning, quantify, the mainstream model of knowledge distilled compression algorithm.
3, to provide multi-backbone training now includes Darknet-YOLOv3, Tiny-YOLOv3, Mobilenetv3-YOLOv3.
[Currently] support functions
- Normal training
- tiny training
- mobilenetv3 training
- Dior training data set
- bdd100k training data set
- visdrone training data set
- Sparse Training
- Normal pruning
- Regular pruning
- Limit pruning (shortcut)
- Tiny pruning
- BNN quantify
- BWN quantify
- quantization stage-wise in layers
- Knowledge distillation
https://github.com/SpursLipu/YOLOv3-ModelCompression-MultidatasetTraining-Multibackbone/blob/master/README.md
