Various implementations of classification models using PyTorch framework.
- python==3.9
- torch==1.9.0
- torchvision==0.10.0
# CIFAR100
python main.py --name base --idx 0 -g 0 -j 8 --dataset cifar100 --datapath ../data --arch resnet --layers 20 --batch-size 128 --run-type train --epochs 300 --wd 1e-4 --lr 0.1 --lr-scheduler cosine --step-location batch --print-freq 100
# ImageNet
python main.py --name base --idx 0 -g 0 1 2 3 -j 16 --dataset imagenet --datapath /dataset/ImageNet --arch resnet --layers 18 --batch-size 256 --run-type train --epochs 100 --wd 1e-4 --lr 0.1 --lr-scheduler cosine --step-location batch --print-freq 1000
python main.py --gpu 0 --workers 8 --dataset cifar100 --datapath ../data --run-type train --resume --load logs/resnet-20/cifar100/base/0
python main.py --gpu 0 --workers 8 --dataset cifar100 --datapath ../data --run-type validate --load logs/resnet-20/cifar100/base/0
| Model | CIFAR10 (%) | CIFAR100 (%) | ImageNet (%) |
|---|---|---|---|
| ResNet-20 [1] | - | - | - |
| ResNet-56 | - | 71.56 | - |
| PreActResNet-20 [2] | - | - | - |
| PreActResNet-56 | - | - | - |
| WideResNet | - | - | - |
| ShuffleNetV2 | - | - | - |
| MobileNetV2 | - | - | - |
| ReXNet | - | - | - |
| - | - | - | - |
[1] He, Kaiming, et al. "Deep residual learning for image recognition." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.
[2] He, Kaiming, et al. "Identity mappings in deep residual networks." European conference on computer vision. Springer, Cham, 2016.