is it better way to construct models via pytorch api directly rather than parse from cfg?

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

is it better way to construct models via pytorch api directly rather than parse from cfg? #4

Closed lucasjinreal closed 4 years ago

lucasjinreal commented 4 years ago

Is it possible convert these model construction way from cfg to directly from pytorch API?

HaloTrouvaille commented 4 years ago

I think it is possible. For example, you can change the backbone according to pytorch API such as ResNet. Howerver, I think cfg is clearer. If you have other problems, you can add my wechat: ssswwwyyy111 to discuss.