- All codes are written by Python +
Pytorch
. - All codes include only core parts (e.g. model only) for simplicity.
- All models MUST have input parameter 'n_classes' to specify the number of output classes and reject others. Others should be defined with default values.
- Accordingly, all models should be able to declared in this form. e.g. "model(n_classes=10)"
- depthwise seperable
- mobilenet_v1
- mobilenet_v2
- squeezenet
- squeezenext
- shufflenet
- efficientnet
- resnet
- resnet with bottleneck
- dilated residual network
- inception
- densenet
- vanilla_gan
- dcgan