This is a PyTorch implementation of LSKNet (https://github.com/zcablii/LSKNet) ImageNet1K-Pretrain
Data prepare: ImageNet with the following folder structure.
│imagenet/
├──train/
│ ├── n01440764
│ │ ├── n01440764_10026.JPEG
│ │ ├── n01440764_10027.JPEG
│ │ ├── ......
│ ├── ......
├──val/
│ ├── n01440764
│ │ ├── ILSVRC2012_val_00000293.JPEG
│ │ ├── ILSVRC2012_val_00002138.JPEG
│ │ ├── ......
│ ├── ......
1. Pytorch >= 1.7
2. timm == 0.4.12
We use 8 GPUs for training by default. Run command (It has been writen in train.sh):
MODEL=lsknet_s
DROP_PATH=0.1
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 bash distributed_train.sh 8 /path/to/imagenet \
--model $MODEL -b 128 --lr 1e-3 --drop-path $DROP_PATH --amp --epochs 300 --model-ema