Reversible Vision Transformers

Reversible Vision Transformers

简介

RevViT is initially described in Reversible Vision Tranformers, which introduce the reversible idea into vision transformer, to reduce the GPU memory footprint required for training.

摘要

Show the paper's abstract

We present Reversible Vision Transformers, a memory efficient architecture design for visual recognition. By decoupling the GPU memory footprint from the depth of the model, Reversible Vision Transformers enable memory efficient scaling of transformer architectures. We adapt two popular models, namely Vision Transformer and Multiscale Vision Transformers, to reversible variants and benchmark extensively across both model sizes and tasks of image classification, object detection and video classification. Reversible Vision Transformers achieve a reduced memory footprint of up to 15.5× at identical model complexity, parameters and accuracy, demonstrating the promise of reversible vision transformers as an efficient backbone for resource limited training regimes. Finally, we find that the additional computational burden of recomputing activations is more than overcome for deeper models, where throughput can increase up to 3.9× over their non-reversible counterparts.

使用方式

Predict imageUse the modelTest Command

from mmpretrain import inference_model

predict = inference_model('revvit-small_3rdparty_in1k', 'demo/bird.JPEG')
print(predict['pred_class'])
print(predict['pred_score'])

Models and results

Image Classification on ImageNet-1k

模型	预训练	Params (M)	Flops (G)	Top-1 (%)	Top-5 (%)	配置文件	下载
revvit-small_3rdparty_in1k*	从头训练	22.44	4.58	79.87	94.90	config	model
revvit-base_3rdparty_in1k*	从头训练	87.34	17.49	81.81	95.56	config	model

Models with * are converted from the official repo. The config files of these models are only for inference. We haven’t reproduce the training results.

引用

@inproceedings{mangalam2022reversible,
  title={Reversible Vision Transformers},
  author={Mangalam, Karttikeya and Fan, Haoqi and Li, Yanghao and Wu, Chao-Yuan and Xiong, Bo and Feichtenhofer, Christoph and Malik, Jitendra},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={10830--10840},
  year={2022}
}