A curated list of vision transformer related resources, including survey, paper, source code, etc. Maintainer: Murufeng

We are looking for a maintainer! Let me know if interested.

Please feel free to pull requests or open an issue to add papers and source codes.

• 魔改算子
• Local & Hierarchical & multi-scale
• Transformer+卷积结合
• Transformer模型压缩轻量化处理
• DETR变种

• MLP-Mixer

• ResMLP

• RepMLP

• gMLP

• Spatial shift($S^{2}$) MLP V1

• S^2-MLPv2: Improved Spatial-Shift MLP Architecture for Vision

• ViP(Vision Permute-MLP)

• GFNet | MLP领域再发力，清华大学提出将FFT思想用于空间信息交互

• CycleMLP: A MLP-like Architecture for Dense Prediction Github

• AS-MLP

• 1.目标检测（Object-Detection）
• 2.超分辨率（Super-Resolution）
• 3.图像分割、语义分割(Segmentation)
• 4.GAN/生成式/对抗式(GAN/Generative/Adversarial)
• 5.track
• 6.video
• 7.多模态结合
• 8.人体姿态估计
• 9.神经网络架构搜索NAS
• 10.人脸识别
• 11.行人重识别
• 12.密集人群检测
• 13.医学图像处理
• 14.图像风格迁移
• 15.low level vision(去噪，去雨，复原，去模糊等等)

• MLP
• Attention
• Transformer

• A Survey of Transformers
  • 论文作者&单位: 复旦大学邱锡鹏团队; Tianyang Lin, Yuxin Wang, Xiangyang Liu, Xipeng Qiu
  • 时间: 2021.6.08
• A Survey on Visual Transformer
  • 论文作者&单位: 华为诺亚方舟; Kai Han, Yunhe Wang, Hanting Chen, etc
  • 2021.1.30
• Transformers in Vision: A Survey
  • 论文作者:Salman Khan, Muzammal Naseer, Munawar Hayat, Syed Waqas Zamir, Fahad Shahbaz Khan, Mubarak Shah
  • 时间: 2021.1.4

• 一文总结微软研究院Transformer霸榜模型三部曲
• 颜水成团队提出VOLO屠榜CV任务，无需额外训练数据，首次在ImageNet 上达到87.1%
• 分层级联Transformer！苏黎世联邦提出TransCNN: 显著降低了计算/空间复杂度！
• 登上更高峰！颜水成、程明明团队开源ViP，引入三维信息编码机制，无需卷积与注意力
• 清华鲁继文团队提出DynamicViT：一种高效的动态稀疏化Token的ViT
• 并非所有图像都值16x16个词--- 清华&华为提出一种自适应序列长度的动态ViT
• 注意力可以使MLP完全替代CNN吗？ 未来有哪些研究方向？
• 超越Swin Transformer！谷歌提出了收敛更快、鲁棒性更强、性能更强的NesT

• Visual Parser: Representing Part-whole Hierarchies with Transformers -- 牛津大学
• Crossformer: A versatile vision Transformer based on cross-scale attention 浙大&哥大
• Contextual Transformer Networks for Visual Recognition - 京东AI梅涛团队
• Rethinking and Improving Relative Position Encoding for Vision Transformer - 中山大学&微软
• Local-to-Global Self-Attention in Vision Transformers -Inception Institute of Artificial Intelligence(IIAI)
• Polarized Self-Attention: Towards High-quality Pixel-wise Regression -- 新的自注意力机制，屠榜分割任务
• A Unified Efficient Pyramid Transformer for Semantic Segmentation - 李沐 Amazon & 复旦
• 华为诺亚提出CMT：卷积神经网络遇见视觉Transformer
  • paper : CMT: Convolutional Neural Networks Meet Vision Transformers
• 超越Swin，Transformer屠榜三大视觉任务！微软推出新作：Focal Self-Attention paper
• Multi-Scale Densenet续作？搞定Transformer降采样，清华联合华为开源动态ViT！ paper
• CSWin Transformer：具有十字形窗口的视觉Transformer主干 paper
• Faceboo提出 Early Convolutions Help Transformers See Better
• VOLO: Vision Outlooker for Visual Recognition
  • 论文链接: https://arxiv.org/abs/2106.13112
  • 代码地址: https://github.com/sail-sg/volo
  • 作者团队:颜水成，冯佳时 无需任何额外训练数据，首次在ImageNet数据集上实现87.1%精度
• Vision Permutator: A Permutable MLP-Like Architecture for Visual Recognition
  • paper链接：https://arxiv.org/abs/2106.12368
  • 代码地址：https://github.com/Andrew-Qibin/VisionPermutator
• Scaling Vision Transformers
• CAT: Cross Attention in Vision Transformer
• CoAtNet: Marrying Convolution and Attention for All Data Sizes
  • 作者单位：谷歌大佬;Zihang Dai, Hanxiao Liu, Quoc V. Le, Mingxing Tan
• Shuffle Transformer: Rethinking Spatial Shuffle for Vision Transformer
• Container: Context Aggregation Network
• Aggregating Nested Transformers
• X-volution: On the unification of convolution and self-attention
• Video Swin Transformer
• Dynamic Head: Unifying Object Detection Heads with Attentions

• 超越Swin，Transformer屠榜三大视觉任务！微软推出新作：Focal Self-Attention
  • paper
  • code
• **CSWin Transformer：具有十字形窗口的视觉Transformer主干(Swin Transformer 的进阶版) *
  • paper
  • code
• Dynamic Head: Unifying Object Detection Heads with Attentions
  • paper
  • code
• [Swin Transformer] Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
  • paper
  • code

• [VTs]: Visual Transformers: Token-based Image Representation and Processing for Computer Vision]
• [So-ViT]: So-ViT: Mind Visual Tokens for Vision Transformer
  • paper
  • code
• [Token Labeling] Token Labeling: Training a 85.5% Top-1 Accuracy Vision Transformer with 56M Parameters on ImageNet
  • paper
  • code
• [LeViT] LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference]
• [CrossViT] CrossViT: Cross-Attention Multi-Scale Vision Transformer for Image Classification
• [CeiT] Incorporating Convolution Designs into Visual Transformers
• [DeepViT] DeepViT: Towards Deeper Vision Transformer
• [TNT] Transformer in Transformer
  • paper
  • code
• [T2T-ViT] Tokens-to-Token ViT: Training Vision Transformers from Scratch on ImageNet
  • paper
  • code
• [BoTNet] Bottleneck Transformers for Visual Recognition]
• [Visformer] Visformer: The Vision-friendly Transformer
  • paper
  • code
• [ConTNet] ConTNet: Why not use convolution and transformer at the same time?
  • paper
  • code
• [DeiT]: Training data-efficient image transformers & distillation through attention
  • paper
  • code

• [Twins] Twins: Revisiting Spatial Attention Design in Vision Transformers
  • paper
  • code
• Scaling Vision Transformers
• [GasHis-Transformer] GasHis-Transformer: A Multi-scale Visual Transformer Approach for Gastric Histopathology Image Classification
• [Vision Transformer] An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (ICLR)
  • paper
  • code
• [RegionViT] Regional-to-Local Attention for Vision Transformers
• [PVT] Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions]
  • code
• [FPT] Feature Pyramid Transformer (CVPR)
  • paper
  • code
• [PiT] Rethinking Spatial Dimensions of Vision Transformers
  • [paper
  • code
• [CoaT] Co-Scale Conv-Attentional Image Transformers
  • paper
  • code
• [LocalViT] LocalViT: Bringing Locality to Vision Transformers
  • paper
  • code
• [Swin Transformer] Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
  • paper
  • code
• [DPT] Vision Transformers for Dense Prediction
  • paper
  • code
• [MViT] Mask Vision Transformer for Facial Expression Recognition in the wild]

• [Shuffle Transformer] Shuffle Transformer: Rethinking Spatial Shuffle for Vision Transformer
• [TransCNN] Transformer in Convolutional Neural Networks\
  • paper
  • code
• [ResT] ResT: An Efficient Transformer for Visual Recognition
  • paper
  • code
• [CPVT] Do We Really Need Explicit Position Encodings for Vision Transformers?
  • paper
  • code
• [ConViT] ConViT: Improving Vision Transformers with Soft Convolutional Inductive Biases
• [CoaT] Co-Scale Conv-Attentional Image Transformers
  • paper
  • code
• [CvT] CvT: Introducing Convolutions to Vision Transformers
  • paper
  • code
• [ConTNet] ConTNet: Why not use convolution and transformer at the same time?
  • paper
  • code
• [CeiT] Incorporating Convolution Designs into Visual Transformers
• [BoTNet] Bottleneck Transformers for Visual Recognition
• [CPTR] CPTR: Full Transformer Network for Image Captioning

• [DynamicViT]: Efficient Vision Transformers with Dynamic Token Sparsification
  • paper
  • code
• [DVT] Not All Images are Worth 16x16 Words: Dynamic Vision Transformers with Adaptive Sequence Length
• [LeViT] LeViT: a Vision Transformer in ConvNet's Clothing for Faster Inference

• [UP-DETR] UP-DETR: Unsupervised Pre-training for Object Detection with Transformers (CVPR)
• [Deformable DETR] Deformable DETR: Deformable Transformers for End-to-End Object Detection (ICLR)
  • paper
  • code
• [DETR] End-to-End Object Detection with Transformers (ECCV)
  • paper
  • code
• [Meta-DETR] Meta-DETR: Few-Shot Object Detection via Unified Image-Level Meta-Learning
  • paper
  • code
• [DA-DETR] DA-DETR: Domain Adaptive Detection Transformer by Hybrid Attention
• [DETReg] Unsupervised Pretraining with Region Priors for Object Detection
  • paper
  • code

• [Pointformer] 3D Object Detection with Pointformer
• [ViT-FRCNN] Toward Transformer-Based Object Detection
• Oriented Object Detection with Transformer
• [YOLOS] You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection
  • paper
  • code
• [COTR] COTR: Convolution in Transformer Network for End to End Polyp Detection
• [TransVOD] End-to-End Video Object Detection with Spatial-Temporal Transformers
  • paper
  • code
• [CAT] CAT: Cross-Attention Transformer for One-Shot Object Detection
• [M2TR] M2TR: Multi-modal Multi-scale Transformers for Deepfake Detection
• Transformer Transforms Salient Object Detection and Camouflaged Object Detection
• [SSTN] SSTN: Self-Supervised Domain Adaptation Thermal Object Detection for Autonomous Driving
• [TSP-FCOS] Rethinking Transformer-based Set Prediction for Object Detection
• [ACT] End-to-End Object Detection with Adaptive Clustering Transformer
• [PED] DETR for Pedestrian Detection
• [DPT] Vision Transformers for Dense Prediction
  • paper
  • code

• [TTSR] Learning Texture Transformer Network for Image Super-Resolution (CVPR)
  • paper
  • code

• Fully Transformer Networks for Semantic ImageSegmentation
• [TransVOS] TransVOS: Video Object Segmentation with Transformers
• [SegFormer] SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers
  • paper
  • code
• [VisTR] End-to-End Video Instance Segmentation with Transformers (CVPR)
• [Trans2Seg] Segmenting Transparent Object in the Wild with Transformer
  • paper
  • code
• SETR ：Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers
  • 作者单位：复旦, 牛津大学, 萨里大学, 腾讯优图, Facebook
  • 主页：https://fudan-zvg.github.io/SETR/
  • 代码：https://github.com/fudan-zvg/SETR
  • 论文：https://arxiv.org/abs/2012.15840

• [GANsformer] Generative Adversarial Transformers

  • 链接
  • code

• [TransGAN]: Two Transformers Can Make One Strong GAN

  • 链接
  • Code

• [AOT-GAN] Aggregated Contextual Transformations for High-Resolution Image Inpainting

  • paper
  • code

• [STGT] Spatial-Temporal Graph Transformer for Multiple Object Tracking
• Transformer Tracking
• [TransCenter] TransCenter: Transformers with Dense Queries for Multiple-Object Tracking
• [TrackFormer] TrackFormer: Multi-Object Tracking with Transformers
• [TransTrack] TransTrack: Multiple-Object Tracking with Transformer
  • paper
  • code

• Video Swin Transformer
• Anticipative Video Transformer
  • paper
  • code
• [TimeSformer] Is Space-Time Attention All You Need for Video Understanding?
• [VidTr] VidTr: Video Transformer Without Convolutions
• [ViViT] ViViT: A Video Vision Transformer
• [VTN] Video Transformer Network
• [VisTR] End-to-End Video Instance Segmentation with Transformers (CVPR)
• [STTN] Learning Joint Spatial-Temporal Transformations for Video Inpainting (ECCV)
  • paper
  • code

• ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision
  • 论文：https://arxiv.org/abs/2102.03334
  • 代码：https://github.com/dandelin/ViLT

• [TransPose] TransPose: Towards Explainable Human Pose Estimation by Transformer
• [TFPose] TFPose: Direct Human Pose Estimation with Transformers
• Lifting Transformer for 3D Human Pose Estimation in Video

• [BossNAS] BossNAS: Exploring Hybrid CNN-transformers with Block-wisely Self-supervised Neural Architecture Search
  • paper
  • code
• Vision Transformer Architecture Search

• [FaceT]: Learning to Cluster Faces via Transformer

• [TransReID] TransReID: Transformer-based Object Re-Identification

• [TransCrowd] TransCrowd: Weakly-Supervised Crowd Counting with Transformer
  • paper
  • code

• [SUNETR] SUNETR: Transformers for 3D Medical Image Segmentation]
• [U-Transformer] U-Net Transformer: Self and Cross Attention for Medical Image Segmentation
• [MedT] Medical Transformer: Gated Axial-Attention for Medical Image Segmentation
  • paper
  • code
• [TransUNet] TransUNet: Transformers Make Strong Encoders for Medical Image Segmentation
  • paper
  • code

• StyTr2: Unbiased Image Style Transfer with Transformers
  • paper

• [IPT] Pre-Trained Image Processing Transformer (CVPR)
• [SDNet]: mutil-branch for single image deraining using swin
  • paper
  • code
• Uformer: A General U-Shaped Transformer for Image Restoration
  • paper
  • code

• Chasing Sparsity in Vision Transformers:An End-to-End Exploration
• MViT: Mask Vision Transformer for Facial Expression Recognition in the wild
• [CPTR] CPTR: Full Transformer Network for Image Captioning
• Learn to Dance with AIST++: Music Conditioned 3D Dance Generation
  • paper
  • code
• Deepfake Video Detection Using Convolutional Vision Transformer
• Training Vision Transformers for Image Retrieval

• MLP-Mixer

• ResMLP

• RepMLP

• gMLP

• Spatial shift($S^{2}) MLP

• ViP(Vision Permute-MLP)

• 待更新

• 待更新