ParFormer: Vision Transformer Baseline with Parallel Local Global Token Mixer and Convolution Attention Patch Embedding

The Implementation of ParFormer: Vision Transformer Baseline with Parallel Local Global Token Mixer and Convolution Attention Patch Embedding

arXiv | PDF

ImageNet

Model Name	Resolution	Params	GFLOPs	@Top-1	Download
ParFormer-B1	224X224	11M	1.5	80.5	model
ParFormer-B2	224X224	23M	3.4	82.1	model
ParFormer-B3	224X224	34M	6.5	83.1	model

Prerequisites

conda virtual environment is recommended.

conda install pytorch torchvision cudatoolkit=11.8 -c pytorch
pip install timm==0.6.13
pip install wandb
pip install fvcore

Data preparation

Download and extract ImageNet train and val images from http://image-net.org/. The training and validation data are expected to be in the train folder and val folder respectively:

|-- /path/to/imagenet/
    |-- train
    |-- val

Single machine multi-GPU training

We provide an example training script train_imnet.sh using PyTorch distributed data parallel (DDP).

To train ParFormer-B1 on an 2-GPU machine:

sh train_imnet.sh parformer_b1 2

Tips: specify your data path and experiment name in the script!