PatchDropout: Economizing Vision Transformers Using Patch Dropout

Yue Liu, Christos Matsoukas, Fredrik Strand, Hossein Azizpour, Kevin Smith; Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2023, pp. 3953-3962


Vision transformers have demonstrated the potential to outperform CNNs in a variety of vision tasks. But the computational and memory requirements of these models prohibit their use in many applications, especially those that depend on high-resolution images, such as medical image classification. Efforts to train ViTs more efficiently are overly complicated, necessitating architectural changes or intricate training schemes. In this work, we show that standard ViT models can be efficiently trained at high resolution by randomly dropping input image patches. This simple approach, PatchDropout, reduces FLOPs and memory by at least 50% in standard natural image datasets such as ImageNet, and those savings only increase with image size. On CSAW, a high-resolution medical dataset, we observe a 5x savings in computation and memory using PatchDropout, along with a boost in performance. For practitioners with a fixed computational or memory budget, PatchDropout makes it possible to choose image resolution, hyperparameters, or model size to get the most performance out of their model.

Related Material

[pdf] [supp] [arXiv]
@InProceedings{Liu_2023_WACV, author = {Liu, Yue and Matsoukas, Christos and Strand, Fredrik and Azizpour, Hossein and Smith, Kevin}, title = {PatchDropout: Economizing Vision Transformers Using Patch Dropout}, booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)}, month = {January}, year = {2023}, pages = {3953-3962} }