PyTorch Implementation of ResNet-preact

Requirements

• Python >= 3.6
• PyTorch >= 1.0.1
• torchvision
• tensorboardX
• YACS

Usage

$ python train.py model.block_type basic model.depth 110 run.outdir results

Use PyramidNet-like Residual Unit

$ python train.py model.block_type basic model.depth 110 model.remove_first_relu True model.add_last_bn True run.outdir results

Results on CIFAR-10

Model	Test Error (median of 3 runs)	Test Error (in paper)	Training Time
ResNet-preact-110	6.47	6.37 (median of 5 runs)	3h05m
ResNet-preact-164 bottleneck	5.90	5.46 (median of 5 runs)	4h01m

References

• He, Kaiming, et al. "Deep residual learning for image recognition." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016. arXiv:1512.03385
• He, Kaiming, et al. "Identity mappings in deep residual networks." European Conference on Computer Vision. Springer International Publishing, 2016. arXiv:1603.05027, Torch implementation