search

HaloTrouvaille / YOLO-Multi-Backbones-Attention

Model Compression—YOLOv3 with multi lightweight backbones(ShuffleNetV2 HuaWei GhostNet), attention, prune and quantization
https://github.com/HaloTrouvaille/YOLO-Multi-Backbones-Attention
493 stars 118 forks source link

量化问题 #10

Closed yolunghiu closed 4 years ago

yolunghiu commented 4 years ago

您好, 代码里量化时没有对BN层做处理, 参数仍然是fp32, 这个在部署时要怎么处理呢?

HaloTrouvaille commented 4 years ago

您好，后续会更新融合BN的量化。