Open cloveropen opened 2 years ago
The get started tutorial is at https://github.com/facebookresearch/d2go/tree/main/demo#quantization-aware-training
Example config is at https://github.com/facebookresearch/d2go/blob/main/configs/qat_faster_rcnn_fbnetv3a_C4.yaml
when trained a model,the accuracy is fairly good,but when convert to pt with torchscript_int8,the Accuracy becomes very poor, I found Quantization aware training ,but now document about it ,can anyone give ways to improve accuracy?