Vision-Transformer-papers

This repository contains a (non-exhaustive) overview of follow-up works based on the original Vision Transformer (ViT) by Google. Feel free to open a PR to add more papers!

Distillation:

• DeiT (Data-efficient Image Transformers): https://arxiv.org/abs/2012.12877
• Efficient Vision Transformers via Fine-Grained Manifold Distillation: https://arxiv.org/abs/2107.01378
• NViT (Vision Transformer Compression and Parameter Redistribution): https://arxiv.org/abs/2110.04869
• SiT (Self-slimmed Vision Transformer): https://arxiv.org/abs/2111.12624

New pre-training objectives:

• self-supervised:
  • DINO (Emerging Properties in Self-Supervised Vision Transformers): https://arxiv.org/abs/2104.14294
  • MoBY (Self-Supervised Learning with Swin Transformers): https://arxiv.org/abs/2105.04553
  • EsViT (Efficient self-supervised Vision Transformers): https://arxiv.org/abs/2106.09785
  • BEiT (BERT Pre-Training of Image Transformers): https://arxiv.org/abs/2106.08254
  • MAE (Masked Autoencoders Are Scalable Vision Learners): https://arxiv.org/abs/2111.06377
  • SiT (Self-supervised vIsion Transformer): https://arxiv.org/abs/2104.03602
  • SimMIM (A Simple Framework for Masked Image Modeling): https://arxiv.org/abs/2111.09886
• supervised:
  • Token Labeling for Better Training Vision Transformers: https://arxiv.org/abs/2104.10858
  • Vision Transformers with Patch Diversification: https://arxiv.org/abs/2104.12753
  • Token Pooling in Vision Transformers: https://arxiv.org/abs/2110.03860

New pre-training tricks, techniques:

• Scaling Vision Transformers: https://arxiv.org/abs/2106.04560
• Vision Transformers with Patch Diversification: https://arxiv.org/abs/2104.12753
• Billion-Scale Pretraining with Vision Transformers for Multi-Task Visual Representations: https://arxiv.org/abs/2108.05887
• How to train your ViT? Data, Augmentation, and Regularization in Vision Transformers: https://arxiv.org/abs/2106.10270
• When Vision Transformers Outperform ResNets without Pretraining or Strong Data Augmentations (SAM optimizer): https://arxiv.org/abs/2106.01548
• V-MoE (Scaling Vision with Sparse Mixture of Experts): https://arxiv.org/abs/2106.05974

Architectural changes:

• Combining convolution with self-attention:
  • CvT (Introducing convolutions to Vision Transformers): https://arxiv.org/abs/2103.15808
  • ConViT (Improving Vision Transformers with Soft Convolutional Inductive Biases): https://arxiv.org/abs/2103.10697
  • CMT (Convolutional Neural Networks Meet Vision Transformers): https://arxiv.org/abs/2107.06263
  • LeViT (A Vision Transformer in ConvNet's Clothing for Faster Inference): https://arxiv.org/abs/2104.01136
  • Co-Scale Conv-Attentional Image Transformers (CoaT): https://arxiv.org/abs/2104.06399
  • Visformer (The Vision-friendly Transformer): https://arxiv.org/abs/2104.12533
  • CCT (Escaping the Big Data Paradigm with Compact Transformers): https://arxiv.org/abs/2104.05704
  • Refiner (Refining Self-attention for Vision Transformers): https://arxiv.org/abs/2106.03714
  • LVT (Lite Vision Transformer with Enhanced Self-Attention): https://arxiv.org/abs/2112.10809
• Others:
  • PiT (Rethinking Spatial Dimensions of Vision Transformers): https://arxiv.org/abs/2103.16302
  • xCiT (Cross-Covariance Image Transformer): https://arxiv.org/abs/2106.09681
  • EsViT (Efficient self-supervised Vision Transformers): https://arxiv.org/abs/2106.09785
  • Token-to-token ViT (Training ViT from scratch on ImageNet): https://arxiv.org/abs/2101.11986
  • DeepViT (Towards Deeper Vision Transformer): https://arxiv.org/abs/2103.11886
  • PVT (Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions): https://arxiv.org/abs/2102.12122
  • PVTv2 (Improved Baselines with Pyramid Vision Transformer): https://arxiv.org/abs/2106.13797
  • Wider Vision Transformer (Go Wider Instead of Deeper): https://arxiv.org/abs/2107.11817
  • CaiT (Going Deeper with Image Transformers): https://arxiv.org/abs/2103.17239
  • CrossViT (Cross-Attention Multi-Scale Vision Transformer for Image Classification): https://arxiv.org/abs/2103.14899
  • Twins-CVT (Spatial Attention in Vision Transformers): https://arxiv.org/abs/2104.13840
  • LIT (Less is More: Pay Less Attention in Vision Transformers): https://arxiv.org/abs/2105.14217
  • TnT (Transformer-in-Transformer): https://arxiv.org/abs/2103.00112
  • Dynamic Vision Transformer: https://arxiv.org/abs/2105.15075
  • Swin Transformer (Hierarchical Vision Transformer using Shifted Windows): https://arxiv.org/abs/2103.14030
  • Shuffle Transformer (Rethinking Spatial Shuffle for Vision Transformer): https://arxiv.org/abs/2106.03650
  • NesT (Aggregating Nested Transformers): https://arxiv.org/abs/2105.12723
  • Long-Short Transformer (Efficient Transformers for Language and Vision): https://t.co/V8qKUkVH1c?amp=1
  • DynamicViT (Efficient Vision Transformers with Dynamic Token Sparsification): https://arxiv.org/abs/2106.02034
  • Dynamic Transformer (Dynamic Vision Transformers with Adaptive Sequence Length): https://arxiv.org/abs/2105.15075
  • PS-ViT (Vision Transformer with Progressive Sampling): https://arxiv.org/abs/2108.01684
  • RegionViT (Regional-to-Local Attention for Vision Transformers): https://arxiv.org/abs/2106.02689
  • Focal Transformer (Focal Self-attention for Local-Global Interactions in Vision Transformers): https://arxiv.org/pdf/2107.00641.pdf
  • kVT (k-NN Attention for Boosting Vision Transformers): https://arxiv.org/abs/2106.00515
  • Robust Vision Transformer: https://arxiv.org/abs/2105.07926
  • Glance-and-Gaze Vision Transformer: https://arxiv.org/abs/2106.02277
  • Feature Fusion Vision Transformer: https://arxiv.org/abs/2107.02341
  • Augmented Shortcuts for Vision Transformers: https://arxiv.org/abs/2106.15941
  • CrossFormer (A Versatile Vision Transformer Based on Cross-scale Attention): https://arxiv.org/abs/2108.00154
  • CSWin Transformer (A General Vision Transformer Backbone with Cross-Shaped Windows): https://arxiv.org/pdf/2107.00652.pdf
  • Evo-ViT (Slow-Fast Token Evolution for Dynamic Vision Transformer): https://arxiv.org/abs/2108.01390
  • PSViT (Better Vision Transformer via Token Pooling and Attention Sharing): https://t.co/OOnONItfnX?amp=1
  • ImageRPE (relative position encodings) for Vision Transformers: https://arxiv.org/abs/2107.14222
  • What makes for Hierarchical Vision Transformer? https://arxiv.org/abs/2107.02174
  • Multi-Scale Vision Longformer: https://arxiv.org/abs/2103.15358
  • CSWin Transformer: https://arxiv.org/abs/2107.00652
  • MetaFormer is Actually What You Need for Vision: https://arxiv.org/abs/2111.11418
  • Stochastic Layers in Vision Transformers: https://arxiv.org/abs/2112.15111
  • ViR: the Vision Reservoir: https://arxiv.org/abs/2112.13545
  • Blending Anti-Aliasing into Vision Transformer: https://arxiv.org/abs/2110.15156
  • ELSA: Enhanced Local Self-Attention for Vision Transformer: https://arxiv.org/abs/2112.12786
  • Swin Transformer V2: Scaling Up Capacity and Resolution: https://arxiv.org/abs/2111.09883

Investigations of the inner workings (cfr. BERTology):

• Are Convolutional Neural Networks or Transformers more like human vision? https://arxiv.org/abs/2105.07197
• Do Vision Transformers See Like Convolutional Neural Networks? https://arxiv.org/abs/2108.08810
• What makes for Hierarchical Vision Transformer? (Survey on Swin + ShuffleTransformer): https://arxiv.org/abs/2107.02174
• Intriguing Properties of Vision Transformers: https://arxiv.org/abs/2105.10497
• Exploring Corruption Robustness: Inductive Biases in Vision Transformers and MLP-Mixers: https://arxiv.org/abs/2106.13122
• Are Transformers More Robust Than CNNs? https://arxiv.org/abs/2111.05464v1

Applying ViT to other domains besides image classification:

• YOLOS (object detection): https://arxiv.org/abs/2106.00666
• ViTGAN (GANs): https://arxiv.org/abs/2107.04589
• SegFormer (semantic segmentation): https://arxiv.org/abs/2105.15203
• Feature Fusion Vision Transformer (Fine-Grained Visual Categorization): https://arxiv.org/abs/2107.02341
• TrOCR (optical character recognition): https://arxiv.org/abs/2109.10282