Training Vision Transformers with Only 2040 Images
This addresses the data-hungry nature of ViTs for researchers and practitioners working with small datasets, though it is incremental as it builds on existing ViT methods.
The paper tackles the problem of training Vision Transformers (ViTs) with limited data, such as only 2040 images, by proposing a method based on parametric instance discrimination, achieving state-of-the-art results on 7 small datasets and showing that these representations can improve large-scale ImageNet training.
Vision Transformers (ViTs) is emerging as an alternative to convolutional neural networks (CNNs) for visual recognition. They achieve competitive results with CNNs but the lack of the typical convolutional inductive bias makes them more data-hungry than common CNNs. They are often pretrained on JFT-300M or at least ImageNet and few works study training ViTs with limited data. In this paper, we investigate how to train ViTs with limited data (e.g., 2040 images). We give theoretical analyses that our method (based on parametric instance discrimination) is superior to other methods in that it can capture both feature alignment and instance similarities. We achieve state-of-the-art results when training from scratch on 7 small datasets under various ViT backbones. We also investigate the transferring ability of small datasets and find that representations learned from small datasets can even improve large-scale ImageNet training.