This folder contains the implementation of the MoBY with Swin Transformer for image classification.

• ¹ denotes the result of MoBY which has adopted a trick from MoCo v3 that replace theLayerNorm layers before the MLP blocks by BatchNorm.

• Clone this repo:

git clone https://github.com/Swin-Transformer/Transformer-SSL
cd Transformer-SSL

• Create a conda virtual environment and activate it:

conda create -n transformer-ssl python=3.7 -y
conda activate transformer-ssl

• Install CUDA==10.1 with cudnn7 following the official installation instructions
• Install PyTorch==1.7.1 and torchvision==0.8.2 with CUDA==10.1:

conda install pytorch==1.7.1 torchvision==0.8.2 cudatoolkit=10.1 -c pytorch

• Install timm==0.3.2:

pip install timm==0.3.2

• Install Apex:

git clone https://github.com/NVIDIA/apex
cd apex
pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./

• Install other requirements:

pip install opencv-python==4.4.0.46 termcolor==1.1.0 yacs==0.1.8 diffdist

We use standard ImageNet dataset, you can download it from http://image-net.org/. We provide the following two ways to load data:

• For standard folder dataset, move validation images to labeled sub-folders. The file structure should look like:

  $ tree data
  imagenet 
  ├── train
  │   ├── class1
  │   │   ├── img1.jpeg
  │   │   ├── img2.jpeg
  │   │   └── ...
  │   ├── class2
  │   │   ├── img3.jpeg
  │   │   └── ...
  │   └── ...
  └── val
      ├── class1
      │   ├── img4.jpeg
      │   ├── img5.jpeg
      │   └── ...
      ├── class2
      │   ├── img6.jpeg
      │   └── ...
      └── ...
  

• To boost the slow speed when reading images from massive small files, we also support zipped ImageNet, which includes four files:

  • train.zip, val.zip: which store the zipped folder for train and validate splits.
  • train_map.txt, val_map.txt: which store the relative path in the corresponding zip file and ground truth label. Make sure the data folder looks like this:

  $ tree data
  data
  └── ImageNet-Zip
      ├── train_map.txt
      ├── train.zip
      ├── val_map.txt
      └── val.zip
  
  $ head -n 5 data/ImageNet-Zip/val_map.txt
  ILSVRC2012_val_00000001.JPEG	65
  ILSVRC2012_val_00000002.JPEG	970
  ILSVRC2012_val_00000003.JPEG	230
  ILSVRC2012_val_00000004.JPEG	809
  ILSVRC2012_val_00000005.JPEG	516
  
  $ head -n 5 data/ImageNet-Zip/train_map.txt
  n01440764/n01440764_10026.JPEG	0
  n01440764/n01440764_10027.JPEG	0
  n01440764/n01440764_10029.JPEG	0
  n01440764/n01440764_10040.JPEG	0
  n01440764/n01440764_10042.JPEG	0

To train MoBY with Swin Transformer on ImageNet, run:

python -m torch.distributed.launch --nproc_per_node <num-of-gpus-to-use> --master_port 12345 moby_main.py \ 
--cfg <config-file> --data-path <imagenet-path> [--batch-size <batch-size-per-gpu> --output <output-directory> --tag <job-tag>]

• Recommand using --output and --tag to tidy your experiments.

Notes:

• To use zipped ImageNet instead of folder dataset, add --zip to the parameters.
  • To cache the dataset in the memory instead of reading from files every time, add --cache-mode part, which will shard the dataset into non-overlapping pieces for different GPUs and only load the corresponding one for each GPU.
• When GPU memory is not enough, you can try the following suggestions:
  • Use gradient accumulation by adding --accumulation-steps <steps>, set appropriate <steps> according to your need.
  • Use gradient checkpointing by adding --use-checkpoint, e.g., it saves about 60% memory when training Swin-B. Please refer to this page for more details.
  • We recommend using multi-node with more GPUs for training very large models, a tutorial can be found in this page.
• To change config options in general, you can use --opts KEY1 VALUE1 KEY2 VALUE2, e.g., --opts TRAIN.EPOCHS 100 TRAIN.WARMUP_EPOCHS 5 will change total epochs to 100 and warm-up epochs to 5.
• For additional options, see config and run python moby_main.py --help to get detailed message.

For example, to train MoBY with Vision Transformers with 8 GPU on a single node for 300 epochs, run:

MoBY Swin-T:

python -m torch.distributed.launch --nproc_per_node 8 --master_port 12345  moby_main.py \
--cfg configs/moby_swin_tiny.yaml --data-path <imagenet-path> --batch-size 64

MoBY DeiT-Small

python -m torch.distributed.launch --nproc_per_node 8 --master_port 12345  moby_main.py \
--cfg configs/moby_deit_small.yaml --data-path <imagenet-path> --batch-size 64

To evaluate a pre-trained MoBY with Swin Transformer on ImageNet-1K linear evaluation, run:

python -m torch.distributed.launch --nproc_per_node <num-of-gpus-to-use> --master_port 12345 moby_linear.py \
--cfg <config-file> --data-path <imagenet-path>

Notes:

• Make sure the <config-file>, <output-directory> and <tag> are the same as in the pre-training stage.
• Note that some configurations are fixed in moby_linear.py for simplicity.

For example, to evaluate MoBY Swin-T with 8 GPU on a single node on ImageNet-1K linear evluation, run:

python -m torch.distributed.launch --nproc_per_node 8 --master_port 12345  moby_linear.py \
--cfg configs/moby_swin_tiny.yaml --data-path <imagenet-path> --batch-size 64

To evaluate a MoBY with Swin Transformer linear evaluation model on ImageNet-1K, run:

python -m torch.distributed.launch --nproc_per_node <num-of-gpus-to-use> --master_port 12345 moby_linear.py \
--cfg <config-file> --resume <checkpoint> --data-path <imagenet-path> --eval

For example, to evaluate the provided MoBY Swin-T linear evaluation model with a single GPU:

python -m torch.distributed.launch --nproc_per_node 1 --master_port 12345 moby_linear.py \
--cfg configs/moby_swin_tiny.yaml --resume moby_swin_t_300ep_linear.pth --data-path <imagenet-path> --eval