Implementations of deep learning models to challenge cifar10 dataset, using PyTorch. The implementation for network traing includes some kind of fancy tools, like prefetch_generator, tqdm and tensorboardx. I also use logging to print information into log file rather than print function.

update May 17th, 2019: I tried the augmentation found by AutoAugment, while still using origin model architecture. And I got 0.94 accuracy.

update May 24th, 2019: I've got 0.96 accuracy on cifar10. I think it's enough. This repo then will be used to keep track about my ideas, my fun of implementing interesting architectures.

• Hardware: 2 GPUs
• Software: Pytorch, prefetch_generator, tensorboardX, tqdm

I only tested ResNet with 110 layers. I will keep update for better performance and for other architechture.

• Improve test accuracy
• Test other architecture