CVPR 2022 全面盘点：最新350篇论文分方向汇总 / 代码 /

转载自CVPR 2022 综合盘点:最新350篇论文分方向汇总: / 代码 / 解读 / 直播 / 项目(更新) - 知乎

CVPR 2022 已上榜，共收到2067篇论文，收到的论文数量比去年增加了24%。CVPR在2022年正式会议之前，为了让大家更快地获得和学习计算机视觉前沿技术，极市对CVPR022 跟踪最新论文，包括论文分为研究方向和代码汇总以及现场分享论文技术。

官网链接：http://CVPR2022.thecvf.com 会议时间：2021年6月19日至6月24日 相关问题：如何评价 CVPR2022 论文接收结果？ 相关报道：CVPR 2022 接收结果出炉！ 2067 接收量增加了24%

此前我们对CVPR2021/CVPR整理了2020/2019/2018，所有内容都总结在我们身上Github：

https://github.com/extreme-assistant/CVPR2022-Paper-Code-Interpretation

github.com/extreme-assistant/CVPR2022-Paper-Code-Interpretation

update: 2022/3/3 更新 19 篇 2022/3/4 更新 29 篇 2022/3/7 更新 17 篇 2022/3/9 更新 57 篇 2022/3/10 更新 8 篇 2022/3/11 更新 18 篇 2022/3/14 更新 11 篇 2022/3/15 更新 30 篇 2022/3/16 更新 16 篇 2022/3/17 更新 24 篇 2022/3/18 更新 25 篇 2022/3/22 更新 52 篇 2022/3/23 更新 29 篇 2022/3/24 更新 22 篇

目录

1. CVPR2022 接受论文/代码分方向汇总(更新) 2. CVPR2022 Oral（更新中） 3. CVPR2022 论文解读总结(更新) 4. CVPR2022 分享极市论文 5. To do list

1.CVPR2022年接受论文/代码分方向整理(持续更新)

点击左侧（PC 可)或顶部(移动端)可跳转到相应类别的论文。

检测

2D目标检测(2D Object Detection)

[12] Progressive End-to-End Object Detection in Crowded Scenes(在拥挤场景中逐步端到端对象检测)paper | code

[11] Real-time Object Detection for Streaming Perception(实时检测流感知对象)paper | code

[10] Oriented RepPoints for Aerial Object Detection(对空中目标进行检测 RepPoints)(小目标检测)paper | code

[9] Confidence Propagation Cluster: Unleash Full Potential of Object Detectors(信心传播集群:释放物体探测器的所有潜力)paper

[8] Semantic-aligned Fusion Transformer for One-shot Object Detection(语义对齐融合转换器用于一次性目标检测)paper

[7] A Dual Weighting Label Assignment Scheme for Object Detection(目标检测双重加权标签分配方案)paper | code

[6] MUM : Mix Image Tiles and UnMix Feature Tiles for Semi-Supervised Object Detection(混合图像块和 UnMix 半监督目标检测采用特征块)paper | code

[5] SIGMA: Semantic-complete Graph Matching for Domain Adaptive Object Detection(域自适应对象检测的语义完全图匹配)paper | code

[4] Accelerating DETR Convergence via Semantic-Aligned Matching(加速语义对齐匹配 DETR 收敛)paper | code

[3] Focal and Global Knowledge Distillation for Detectors(蒸馏探测器的焦点和全球知识) keywords: Object Detection, Knowledge Distillationpaper | code

[2] Unknown-Aware Object Detection: Learning What You Don't Know from Videos in the Wild(未知感知对象检测:从野外视频中学习你不知道的东西)paper | code

[1] Localization Distillation for Dense Object Detection(定位蒸馏密集对象检测) keywords: Bounding Box Regression, Localization Quality Estimation, Knowledge Distillationpaper | code 解释:南开程明明团队和天大提出LD：目标检测定位蒸馏

视频目标检测(Video Object Detection)

[1] Unsupervised Activity Segmentation by Joint Representation Learning and Online Clustering(无监督活动通过联合表示学习和在线聚类划分)paper | video

3D目标检测(3D object detection)

[13] TransFusion: Robust LiDAR-Camera Fusion for 3D Object Detection with Transformers(用于 3D 对象检测稳定 LiDAR-Camera Fusion 与 Transformer)paper | code

[12] Not All Points Are Equal: Learning Highly Efficient Point-based Detectors for 3D LiDAR Point Clouds(学习用于 3D LiDAR 基于点的点云高效检测器)paper | code

[11] Sparse Fuse Dense: Towards High Quality 3D Detection with Depth Completion(走向高质量的深度完成 3D 检测)paper

[10] MonoDTR: Monocular 3D Object Detection with Depth-Aware Transformer(使用深度感知 Transformer 的单目 3D 对象检测)paper | code

[9] Voxel Set Transformer: A Set-to-Set Approach to 3D Object Detection from Point Clouds(从点云开始 3D 对象检测的 Set-to-Set 方法)paper | code

[8] VISTA: Boosting 3D Object Detection via Dual Cross-VIew SpaTial Attentionpaper | code

[7] MonoJSG: Joint Semantic and Geometric Cost Volume for Monocular 3D Object Detection(单目 3D 联合语义和几何成本的目标检测)paper | code

[6] DeepFusion: Lidar-Camera Deep Fusion for Multi-Modal 3D Object Detection(用于多模态 3D 激光雷达相机深度集成的目标检测)paper | code

[5] Point Density-Aware Voxels for LiDAR 3D Object Detection(用于 LiDAR 3D 对象检测的点密度感知元素)paper | code

[4] Back t Reality: Weakly-supervised 3D Object Detection with Shape-guided Label Enhancement(带有形状引导标签增强的弱监督 3D 对象检测)paper | code  

[3] Canonical Voting: Towards Robust Oriented Bounding Box Detection in 3D Scenes(在 3D 场景中实现稳健的定向边界框检测)paper | code  

[2] A Versatile Multi-View Framework for LiDAR-based 3D Object Detection with Guidance from Panoptic Segmentation(在全景分割的指导下，用于基于 LiDAR 的 3D 对象检测的多功能多视图框架) keywords: 3D Object Detection with Point-based Methods, 3D Object Detection with Grid-based Methods, Cluster-free 3D Panoptic Segmentation, CenterPoint 3D Object Detectionpaper  

[1] Pseudo-Stereo for Monocular 3D Object Detection in Autonomous Driving(自动驾驶中用于单目 3D 目标检测的伪立体) keywords: Autonomous Driving, Monocular 3D Object Detectionpaper | code  

人物交互检测(HOI Detection)

伪装目标检测(Camouflaged Object Detection)

[2] Implicit Motion Handling for Video Camouflaged Object Detection(视频伪装对象检测的隐式运动处理)paper | dataset  

[1] Zoom In and Out: A Mixed-scale Triplet Network for Camouflaged Object Detection(放大和缩小：用于伪装目标检测的混合尺度三元组网络)paper | code  

旋转目标检测(Rotation Object Detection)

显著性目标检测(Saliency Object Detection)

[2] Bi-directional Object-context Prioritization Learning for Saliency Ranking(显着性排名的双向对象上下文优先级学习)paper | code  

[1] Democracy Does Matter: Comprehensive Feature Mining for Co-Salient Object Detection()paper  

关键点检测(Keypoint Detection)

[1] UKPGAN: A General Self-Supervised Keypoint Detector(一个通用的自监督关键点检测器)paper | code  

车道线检测(Lane Detection)

[2] CLRNet: Cross Layer Refinement Network for Lane Detection(用于车道检测的跨层细化网络)paper  

[1] Rethinking Efficient Lane Detection via Curve Modeling(通过曲线建模重新思考高效车道检测) keywords: Segmentation-based Lane Detection, Point Detection-based Lane Detection, Curve-based Lane Detection, autonomous drivingpaper | code  

边缘检测(Edge Detection)

[1] EDTER: Edge Detection with Transformer(使用transformer的边缘检测)paper | code  

消失点检测(Vanishing Point Detection)

[1] Deep vanishing point detection: Geometric priors make dataset variations vanish(深度消失点检测：几何先验使数据集变化消失)paper | code  

异常检测(Anomaly Detection)

[4] UBnormal: New Benchmark for Supervised Open-Set Video Anomaly Detection(监督开放集视频异常检测的新基准)paper | code  

[3] ViM: Out-Of-Distribution with Virtual-logit Matching(具有虚拟 logit 匹配的分布外)(OOD检测)paper | code  

[2] Generative Cooperative Learning for Unsupervised Video Anomaly Detection(用于无监督视频异常检测的生成式协作学习)paper  

[1] Self-Supervised Predictive Convolutional Attentive Block for Anomaly Detection(用于异常检测的自监督预测卷积注意力块)(论文暂未上传)paper | code  

分割(Segmentation)

图像分割(Image Segmentation)

[3] Learning What Not to Segment: A New Perspective on Few-Shot Segmentation(学习不分割的内容：关于小样本分割的新视角)paper | code  

[2] CRIS: CLIP-Driven Referring Image Segmentation(CLIP 驱动的参考图像分割)paper  

[1] Hyperbolic Image Segmentation(双曲线图像分割)paper  

全景分割(Panoptic Segmentation)

[2] Panoptic SegFormer: Delving Deeper into Panoptic Segmentation with Transformers(使用 Transformers 深入研究全景分割)paper | code  

[1] Bending Reality: Distortion-aware Transformers for Adapting to Panoramic Semantic Segmentation(弯曲现实：适应全景语义分割的失真感知Transformer) keywords: Semantic- and panoramic segmentation, Unsupervised domain adaptation, Transformerpaper | code  

语义分割(Semantic Segmentation)

[16] DynamicEarthNet: Daily Multi-Spectral Satellite Dataset for Semantic Change Segmentation(用于语义变化分割的每日多光谱卫星数据集)paper | data | website  

[15] Class-Balanced Pixel-Level Self-Labeling for Domain Adaptive Semantic Segmentation(用于域自适应语义分割的类平衡像素级自标记)paper | code  

[14] Regional Semantic Contrast and Aggregation for Weakly Supervised Semantic Segmentation(弱监督语义分割的区域语义对比和聚合)paper | code  

[13] Tree Energy Loss: Towards Sparsely Annotated Semantic Segmentation(走向稀疏注释的语义分割)paper | code  

[12] Scribble-Supervised LiDAR Semantic Segmentationpaper |code  

[11] ADAS: A Direct Adaptation Strategy for Multi-Target Domain Adaptive Semantic Segmentation(多目标域自适应语义分割的直接适应策略)paper  

[10] Weakly Supervised Semantic Segmentation by Pixel-to-Prototype Contrast(通过像素到原型对比的弱监督语义分割)paper  

[9] Representation Compensation Networks for Continual Semantic Segmentation(连续语义分割的表示补偿网络)paper | code  

[8] Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels(使用不可靠伪标签的半监督语义分割)paper | code | project  

[7] Weakly Supervised Semantic Segmentation using Out-of-Distribution Data(使用分布外数据的弱监督语义分割)paper | code  

[6] Self-supervised Image-specific Prototype Exploration for Weakly Supervised Semantic Segmentation(弱监督语义分割的自监督图像特定原型探索)paper | code  

[5] Multi-class Token Transformer for Weakly Supervised Semantic Segmentation(用于弱监督语义分割的多类token Transformer)paper | code  

[4] Cross Language Image Matching for Weakly Supervised Semantic Segmentation(用于弱监督语义分割的跨语言图像匹配)paper  

[3] Learning Affinity from Attention: End-to-End Weakly-Supervised Semantic Segmentation with Transformers(从注意力中学习亲和力：使用 Transformers 的端到端弱监督语义分割)paper | code  

[2] ST++: Make Self-training Work Better for Semi-supervised Semantic Segmentation(让自我训练更好地用于半监督语义分割) keywords: Semi-supervised learning, Semantic segmentation, Uncertainty estimationpaper | code  

[1] Class Re-Activation Maps for Weakly-Supervised Semantic Segmentation(弱监督语义分割的类重新激活图)paper | code  

实例分割(Instance Segmentation)

[6] Mask Transfiner for High-Quality Instance Segmentation(用于高质量实例分割的 Mask Transfiner)paper | code  

[5] ContrastMask: Contrastive Learning to Segment Every Thing(对比学习分割每件事)paper  

[4] Discovering Objects that Can Move(发现可以移动的物体)paper | code  

[3] E2EC: An End-to-End Contour-based Method for High-Quality High-Speed Instance Segmentation(一种基于端到端轮廓的高质量高速实例分割方法)paper | code  

[2] Efficient Video Instance Segmentation via Tracklet Query and Proposal(通过 Tracklet Query 和 Proposal 进行高效的视频实例分割)paper  

[1] SoftGroup for 3D Instance Segmentation on Point Clouds(用于点云上的 3D 实例分割) keywords: 3D Vision, Point Clouds, Instance Segmentationpaper | code  

超像素(Superpixel)

视频目标分割(Video Object Segmentation)

[1] Language as Queries for Referring Video Object Segmentation(语言作为引用视频对象分割的查询)paper | code  

抠图(Matting)

密集预测(Dense Prediction)

[1] DenseCLIP: Language-Guided Dense Prediction with Context-Aware Prompting(具有上下文感知提示的语言引导密集预测)paper | code  

视频处理(Video Processing)

[2] Unifying Motion Deblurring and Frame Interpolation with Events(将运动去模糊和帧插值与事件统一起来)paper  

[1] Neural Compression-Based Feature Learning for Video Restoration(用于视频复原的基于神经压缩的特征学习)paper  

视频编辑(Video Editing)

[1] M3L: Language-based Video Editing via Multi-Modal Multi-Level Transformers(M3L：通过多模式多级transformer进行基于语言的视频编辑)paper  

视频生成/视频合成(Video Generation/Video Synthesis)

[2] Depth-Aware Generative Adversarial Network for Talking Head Video Generation(用于说话头视频生成的深度感知生成对抗网络)paper | code  

[1] Show Me What and Tell Me How: Video Synthesis via Multimodal Conditioning(告诉我什么并告诉我如何：通过多模式调节进行视频合成)paper | code  

估计(Estimation)

光流/运动估计(Optical Flow/Motion Estimation)

[2] Global Matching with Overlapping Attention for Optical Flow Estimation(具有重叠注意力的全局匹配光流估计)paper | code  

[1] CamLiFlow: Bidirectional Camera-LiDAR Fusion for Joint Optical Flow and Scene Flow Estimation(用于联合光流和场景流估计的双向相机-LiDAR 融合)paper  

深度估计(Depth Estimation)

[12] Practical Stereo Matching via Cascaded Recurrent Network with Adaptive Correlation(基于自适应相关的级联循环网络的实用立体匹配)paper | project  

[11] Depth Estimation by Combining Binocular Stereo and Monocular Structured-Light(结合双目立体和单目结构光的深度估计)paper | code  

[10] RGB-Depth Fusion GAN for Indoor Depth Completion(用于室内深度完成的 RGB 深度融合 GAN)paper  

[9] Revisiting Domain Generalized Stereo Matching Networks from a Feature Consistency Perspective(从特征一致性的角度重新审视域广义立体匹配网络)paper  

[8] Deep Depth from Focus with Differential Focus Volume(具有不同焦点体积的焦点深度)paper  

[7] ChiTransformer:Towards Reliable Stereo from Cues(从线索走向可靠的立体声)paper  

[6] Rethinking Depth Estimation for Multi-View Stereo: A Unified Representation and Focal Loss(重新思考多视图立体的深度估计：统一表示和焦点损失)paper | code  

[5] ITSA: An Information-Theoretic Approach to Automatic Shortcut Avoidance and Domain Generalization in Stereo Matching Networks(立体匹配网络中自动避免捷径和域泛化的信息论方法) keywords: Learning-based Stereo Matching Networks, Single Domain Generalization, Shortcut Learningpaper  

[4] Attention Concatenation Volume for Accurate and Efficient Stereo Matching(用于精确和高效立体匹配的注意力连接体积) keywords: Stereo Matching, cost volume construction, cost aggregationpaper | code  

[3] Occlusion-Aware Cost Constructor for Light Field Depth Estimation(光场深度估计的遮挡感知成本构造函数)paper | [code](https://github.com/YingqianWang/OACC- Net)  

[2] NeW CRFs: Neural Window Fully-connected CRFs for Monocular Depth Estimation(用于单目深度估计的神经窗口全连接 CRF) keywords: Neural CRFs for Monocular Depthpaper  

[1] OmniFusion: 360 Monocular Depth Estimation via Geometry-Aware Fusion(通过几何感知融合进行 360 度单目深度估计) keywords: monocular depth estimation(单目深度估计),transformerpaper  

人体解析/人体姿态估计(Human Parsing/Human Pose Estimation)

[9] Ray3D: ray-based 3D human pose estimation for monocular absolute 3D localization(用于单目绝对 3D 定位的基于射线的 3D 人体姿态估计)paper | code  

[8] Capturing Humans in Motion: Temporal-Attentive 3D Human Pose and Shape Estimation from Monocular Video(捕捉运动中的人类：来自单目视频的时间注意 3D 人体姿势和形状估计)paper | video  

[7] Physical Inertial Poser (PIP): Physics-aware Real-time Human Motion Tracking from Sparse Inertial Sensors(来自稀疏惯性传感器的物理感知实时人体运动跟踪)paper | project  

[6] Distribution-Aware Single-Stage Models for Multi-Person 3D Pose Estimation(用于多人 3D 姿势估计的分布感知单阶段模型)paper  

[5] MHFormer: Multi-Hypothesis Transformer for 3D Human Pose Estimation(用于 3D 人体姿势估计的多假设transformer)paper | code  

[4] CDGNet: Class Distribution Guided Network for Human Parsing(用于人类解析的类分布引导网络)paper  

[3] Forecasting Characteristic 3D Poses of Human Actions(预测人类行为的特征 3D 姿势)paper | project | video  

[2] Learning Local-Global Contextual Adaptation for Multi-Person Pose Estimation(学习用于多人姿势估计的局部-全局上下文适应) keywords:Top-Down Pose Estimation(从上至下姿态估计), Limb-based Grouping, Direct Regressionpaper  

[1] MixSTE: Seq2seq Mixed Spatio-Temporal Encoder for 3D Human Pose Estimation in Video(用于视频中 3D 人体姿势估计的 Seq2seq 混合时空编码器) keywords：3D Human Pose Estimation, Transformerpaper  

手势估计(Gesture Estimation)

图像处理(Image Processing)

超分辨率(Super Resolution)

[9] Deep Constrained Least Squares for Blind Image Super-Resolution(用于盲图像超分辨率的深度约束最小二乘)paper  

[8] Local Texture Estimator for Implicit Representation Function(隐式表示函数的局部纹理估计器)paper  

[7] A Text Attention Network for Spatial Deformation Robust Scene Text Image Super-resolution(一种用于空间变形鲁棒场景文本图像超分辨率的文本注意网络)paper | code  

[6] Details or Artifacts: A Locally Discriminative Learning Approach to Realistic Image Super-Resolution(一种真实图像超分辨率的局部判别学习方法)paper | code  

[5] Blind Image Super-resolution with Elaborate Degradation Modeling on Noise and Kernel(对噪声和核进行精细退化建模的盲图像超分辨率)paper | code  

[4] Reflash Dropout in Image Super-Resolution(图像超分辨率中的闪退dropout)paper  

[3] Towards Bidirectional Arbitrary Image Rescaling: Joint Optimization and Cycle Idempotence(迈向双向任意图像缩放：联合优化和循环幂等)paper  

[2] HyperTransformer: A Textural and Spectral Feature Fusion Transformer for Pansharpening(用于全色锐化的纹理和光谱特征融合Transformer)paper ｜ code  

[1] HDNet: High-resolution Dual-domain Learning for Spectral Compressive Imaging(光谱压缩成像的高分辨率双域学习) keywords: HSI Reconstruction, Self-Attention Mechanism, Image Frequency Spectrum Analysispaper  

图像复原/图像增强/图像重建(Image Restoration/Image Reconstruction)

[5] Exploring and Evaluating Image Restoration Potential in Dynamic Scenes(探索和评估动态场景中的图像复原潜力)paper  

[4] Come-Closer-Diffuse-Faster: Accelerating Conditional Diffusion Models for Inverse Problems through Stochastic Contraction(通过随机收缩加速逆问题的条件扩散模型)paper  

[3] Mask-guided Spectral-wise Transformer for Efficient Hyperspectral Image Reconstruction(用于高效高光谱图像重建的掩模引导光谱变换器)paper | code  

[2] Restormer: Efficient Transformer for High-Resolution Image Restoration(用于高分辨率图像复原的高效transformer)paper | code  

[1] Event-based Video Reconstruction via Potential-assisted Spiking Neural Network(通过电位辅助尖峰神经网络进行基于事件的视频重建)paper  

图像去阴影/去反射(Image Shadow Removal/Image Reflection Removal)

图像去噪/去模糊/去雨去雾(Image Denoising)

[4] AP-BSN: Self-Supervised Denoising for Real-World Images via Asymmetric PD and Blind-Spot Network(通过非对称 PD 和盲点网络对真实世界图像进行自监督去噪)paper | code  

[3] IDR: Self-Supervised Image Denoising via Iterative Data Refinement(通过迭代数据细化的自监督图像去噪)paper | code  

[2] Blind2Unblind: Self-Supervised Image Denoising with Visible Blind Spots(具有可见盲点的自监督图像去噪)paper | code  

[1] E-CIR: Event-Enhanced Continuous Intensity Recovery(事件增强的连续强度恢复) keywords: Event-Enhanced Deblurring, Video Representationpaper | code  

图像编辑/图像修复(Image Edit/Inpainting)

[5] High-Fidelity GAN Inversion for Image Attribute Editing(用于图像属性编辑的高保真 GAN 反演)paper | code | project  

[4] Style Transformer for Image Inversion and Editing(用于图像反转和编辑的样式transformer)paper | code  

[3] MISF: Multi-level Interactive Siamese Filtering for High-Fidelity Image Inpainting(用于高保真图像修复的多级交互式 Siamese 过滤)paper | code  

[2] HairCLIP: Design Your Hair by Text and Reference Image(通过文本和参考图像设计你的头发) keywords: Language-Image Pre-Training (CLIP), Generative Adversarial Networkspaper | project  

[1] Incremental Transformer Structure Enhanced Image Inpainting with Masking Positional Encoding(增量transformer结构增强图像修复与掩蔽位置编码) keywords: Image Inpainting, Transformer, Image Generationpaper | code  

图像翻译(Image Translation)

[4] Globetrotter: Connecting Languages by Connecting Images(通过连接图像连接语言)paper  

[3] QS-Attn: Query-Selected Attention for Contrastive Learning in I2I Translation(图像翻译中对比学习的查询选择注意)paper | code  

[2] FlexIT: Towards Flexible Semantic Image Translation(迈向灵活的语义图像翻译)paper  

[1] Exploring Patch-wise Semantic Relation for Contrastive Learning in Image-to-Image Translation Tasks(探索图像到图像翻译任务中对比学习的补丁语义关系) keywords: image translation, knowledge transfer,Contrastive learningpaper  

图像质量评估(Image Quality Assessment)

风格迁移(Style Transfer)

[3] Exact Feature Distribution Matching for Arbitrary Style Transfer and Domain Generalization(任意风格迁移和域泛化的精确特征分布匹配)paper | code  

[2] Style-ERD: Responsive and Coherent Online Motion Style Transfer(响应式和连贯的在线运动风格迁移)paper  

[1] CLIPstyler: Image Style Transfer with a Single Text Condition(具有单一文本条件的图像风格转移) keywords: Style Transfer, Text-guided synthesis, Language-Image Pre-Training (CLIP)paper  

人脸(Face)

[5] Cross-Modal Perceptionist: Can Face Geometry be Gleaned from Voices?(跨模态感知者：可以从声音中收集面部几何形状吗？)paper | project  

[4] Portrait Eyeglasses and Shadow Removal by Leveraging 3D Synthetic Data(利用 3D 合成数据去除人像眼镜和阴影)paper | code  

[3] HP-Capsule: Unsupervised Face Part Discovery by Hierarchical Parsing Capsule Network(分层解析胶囊网络的无监督人脸部分发现)paper  

[2] FaceFormer: Speech-Driven 3D Facial Animation with Transformers(FaceFormer：带有transformer的语音驱动的 3D 面部动画)paper | code  

[1] Sparse Local Patch Transformer for Robust Face Alignment and Landmarks Inherent Relation Learning(用于鲁棒人脸对齐和地标固有关系学习的稀疏局部补丁transformer)paper | code  

人脸识别/检测(Facial Recognition/Detection)

[3] Towards Semi-Supervised Deep Facial Expression Recognition with An Adaptive Confidence Margin(具有自适应置信度的半监督深度面部表情识别)paper | code  

[2] Privacy-preserving Online AutoML for Domain-Specific Face Detection(用于特定领域人脸检测的隐私保护在线 AutoML)paper  

[1] An Efficient Training Approach for Very Large Scale Face Recognition(一种有效的超大规模人脸识别训练方法)paper | code  

人脸生成/合成/重建/编辑(Face Generation/Face Synthesis/Face Reconstruction/Face Editing)

[3] FENeRF: Face Editing in Neural Radiance Fields(神经辐射场中的人脸编辑)paper | project  

[2] GCFSR: a Generative and Controllable Face Super Resolution Method Without Facial and GAN Priors(一种没有面部和 GAN 先验的生成可控人脸超分辨率方法)paper  

[1] Sparse to Dense Dynamic 3D Facial Expression Generation(稀疏到密集的动态 3D 面部表情生成) keywords: Facial expression generation, 4D face generation, 3D face modelingpaper  

人脸伪造/反欺骗(Face Forgery/Face Anti-Spoofing)

[4] Self-supervised Learning of Adversarial Example: Towards Good Generalizations for Deepfake Detection(对抗样本的自监督学习：迈向 Deepfake 检测的良好泛化)paper | code  

[3] Domain Generalization via Shuffled Style Assembly for Face Anti-Spoofing(通过 Shuffled Style Assembly 进行域泛化以进行人脸反欺骗)paper | code  

[2] Voice-Face Homogeneity Tells Deepfakepaper | code  

[1] Protecting Celebrities with Identity Consistency Transformer(使用身份一致性transformer保护名人)paper  

目标跟踪(Object Tracking)

[7] Transforming Model Prediction for Tracking(转换模型预测以进行跟踪)paper | code  

[6] MixFormer: End-to-End Tracking with Iterative Mixed Attention(具有迭代混合注意力的端到端跟踪)paper | code  

[5] Unsupervised Domain Adaptation for Nighttime Aerial Tracking(夜间空中跟踪的无监督域自适应)paper | code  

[4] Iterative Corresponding Geometry: Fusing Region and Depth for Highly Efficient 3D Tracking of Textureless Objects(迭代对应几何：融合区域和深度以实现无纹理对象的高效 3D 跟踪)paper | [code](https://github.com/DLR- RM/3DObjectTracking)  

[3] TCTrack: Temporal Contexts for Aerial Tracking(空中跟踪的时间上下文)paper | code  

[2] Beyond 3D Siamese Tracking: A Motion-Centric Paradigm for 3D Single Object Tracking in Point Clouds(超越 3D 连体跟踪：点云中 3D 单对象跟踪的以运动为中心的范式) keywords: Single Object Tracking, 3D Multi-object Tracking / Detection, Spatial-temporal Learning on Point Cloudspaper  

[1] Correlation-Aware Deep Tracking(相关感知深度跟踪)paper  

图像&视频检索/视频理解(Image&Video Retrieval/Video Understanding)

[2] Bridging Video-text Retrieval with Multiple Choice Questions(桥接视频文本检索与多项选择题)paper | code  

[1] BEVT: BERT Pretraining of Video Transformers(视频Transformer的 BERT 预训练) keywords: Video understanding, Vision transformers, Self-supervised representation learning, BERT pretrainingpaper | code  

行为识别/动作识别/检测/分割/定位(Action/Activity Recognition)

[12] How Do You Do It? Fine-Grained Action Understanding with Pseudo-Adverbs(你怎么做呢？ 使用伪副词进行细粒度的动作理解)paper  

[11] E2(GO)MOTION: Motion Augmented Event Stream for Egocentric Action Recognition(用于以自我为中心的动作识别的运动增强事件流)paper  

[10] Look for the Change: Learning Object States and State-Modifying Actions from Untrimmed Web Videos(寻找变化：从未修剪的网络视频中学习对象状态和状态修改操作)paper | code  

[9] DirecFormer: A Directed Attention in Transformer Approach to Robust Action Recognition(鲁棒动作识别的 Transformer 方法中的定向注意)paper  

[8] Self-supervised Video Transformer(自监督视频transformer)paper | code  

[7] Spatio-temporal Relation Modeling for Few-shot Action Recognition(小样本动作识别的时空关系建模)paper | code  

[6] RCL: Recurrent Continuous Localization for Temporal Action Detection(用于时间动作检测的循环连续定位)paper  

[5] OpenTAL: Towards Open Set Temporal Action Localization(走向开放集时间动作定位)paper | code  

[4] End-to-End Semi-Supervised Learning for Video Action Detection(视频动作检测的端到端半监督学习)paper  

[3] Learnable Irrelevant Modality Dropout for Multimodal Action Recognition on Modality-Specific Annotated Videos(模态特定注释视频上多模态动作识别的可学习不相关模态丢失)paper  

[2] Weakly Supervised Temporal Action Localization via Representative Snippet Knowledge Propagation(通过代表性片段知识传播的弱监督时间动作定位)paper | code  

[1] Colar: Effective and Efficient Online Action Detection by Consulting Exemplars(通过咨询示例进行有效且高效的在线动作检测) keywords:Online action detection(在线动作检测)paper  

行人重识别/检测(Re-Identification/Detection)

[1] Cascade Transformers for End-to-End Person Search(用于端到端人员搜索的级联transformer)paper | code  

图像/视频字幕(Image/Video Caption)

[3] Open-Domain, Content-based, Multi-modal Fact-checking of Out-of-Context Images via Online Resources(通过在线资源对上下文外图像进行开放域、基于内容、多模式的事实检查)paper | code  

[2] Hierarchical Modular Network for Video Captioning(用于视频字幕的分层模块化网络)paper | code  

[1] X -Trans2Cap: Cross-Modal Knowledge Transfer using Transformer for 3D Dense Captioning(使用 Transformer 进行 3D 密集字幕的跨模式知识迁移) keywords：Image Captioning and Dense Captioning(图像字幕/密集字幕)；Knowledge distillation(知识蒸馏)；Transformer；3D Vision(三维视觉)paper  

医学影像(Medical Imaging)

[6] DTFD-MIL: Double-Tier Feature Distillation Multiple Instance Learning for Histopathology Whole Slide Image Classification(用于组织病理学全幻灯片图像分类的双层特征蒸馏多实例学习)paper | code  

[5] ACPL: Anti-curriculum Pseudo-labelling for Semi-supervised Medical Image Classification(半监督医学图像分类的反课程伪标签)paper  

[4] Vox2Cortex: Fast Explicit Reconstruction of Cortical Surfaces from 3D MRI Scans with Geometric Deep Neural Networks(使用几何深度神经网络从 3D MRI 扫描中快速显式重建皮质表面)paper | code  

[3] Generalizable Cross-modality Medical Image Segmentation via Style Augmentation and Dual Normalization(通过风格增强和双重归一化的可泛化跨模态医学图像分割)paper | code  

[2] Adaptive Early-Learning Correction for Segmentation from Noisy Annotations(从噪声标签中分割的自适应早期学习校正) keywords: medical-imaging segmentation, Noisy Annotationspaper | code  

[1] Temporal Context Matters: Enhancing Single Image Prediction with Disease Progression Representations(时间上下文很重要：使用疾病进展表示增强单图像预测) keywords: Self-supervised Transformer, Temporal modeling of disease progressionpaper  

文本检测/识别/理解(Text Detection/Recognition/Understanding)

[3] SwinTextSpotter: Scene Text Spotting via Better Synergy between Text Detection and Text Recognition(通过文本检测和文本识别之间更好的协同作用进行场景文本定位)paper | code  

[2] Fourier Document Restoration for Robust Document Dewarping and Recognition(用于鲁棒文档去扭曲和识别的傅里叶文档恢复)paper | code  

[1] XYLayoutLM: Towards Layout-Aware Multimodal Networks For Visually-Rich Document Understanding(迈向布局感知多模式网络，以实现视觉丰富的文档理解)paper  

遥感图像(Remote Sensing Image)

GAN/生成式/对抗式(GAN/Generative/Adversarial)

[12] Subspace Adversarial Training(子空间对抗训练)paper | code  

[11] DTA: Physical Camouflage Attacks using Differentiable Transformation Network(使用可微变换网络的物理伪装攻击)paper | code  

[10] Improving the Transferability of Targeted Adversarial Examples through Object-Based Diverse Input(通过基于对象的多样化输入提高目标对抗样本的可迁移性)paper | code  

[9] Towards Practical Certifiable Patch Defense with Vision Transformer(使用 Vision Transformer 实现实用的可认证补丁防御)paper

[8] Few Shot Generative Model Adaption via Relaxed Spatial Structural Alignment(基于松弛空间结构对齐的小样本生成模型自适应)paper  

[7] Enhancing Adversarial Training with Second-Order Statistics of Weights(使用权重的二阶统计加强对抗训练)paper | code  

[6] Practical Evaluation of Adversarial Robustness via Adaptive Auto Attack(通过自适应自动攻击对对抗鲁棒性的实际评估)paper | code1 | code2  

[5] Frequency-driven Imperceptible Adversarial Attack on Semantic Similarity(对语义相似性的频率驱动的不可察觉的对抗性攻击)paper

[4] Shadows can be Dangerous: Stealthy and Effective Physical-world Adversarial Attack by Natural Phenomenon(阴影可能很危险：自然现象的隐秘而有效的物理世界对抗性攻击)paper  

[3] Protecting Facial Privacy: Generating Adversarial Identity Masks via Style-robust Makeup Transfer(保护面部隐私：通过风格稳健的化妆转移生成对抗性身份面具)paper  

[2] Adversarial Texture for Fooling Person Detectors in the Physical World(物理世界中愚弄人探测器的对抗性纹理)paper  

[1] Label-Only Model Inversion Attacks via Boundary Repulsion(通过边界排斥的仅标签模型反转攻击)paper

图像生成/图像合成(Image Generation/Image Synthesis)

[11] Modulated Contrast for Versatile Image Synthesis(用于多功能图像合成的调制对比度)paper | code  

[10] Attribute Group Editing for Reliable Few-shot Image Generation(属性组编辑用于可靠的小样本图像生成)paper | code  

[9] Text to Image Generation with Semantic-Spatial Aware GAN(使用语义空间感知 GAN 生成文本到图像)paper | code  

[8] Playable Environments: Video Manipulation in Space and Time(可播放环境：空间和时间的视频操作)paper | code  

[7] FLAG: Flow-based 3D Avatar Generation from Sparse Observations(从稀疏观察中生成基于流的 3D 头像)paper | project  

[6] Dynamic Dual-Output Diffusion Models(动态双输出扩散模型)paper  

[5] Exploring Dual-task Correlation for Pose Guided Person Image Generation(探索姿势引导人物图像生成的双任务相关性)paper | code  

[4] 3D Shape Variational Autoencoder Latent Disentanglement via Mini-Batch Feature Swapping for Bodies and Faces(基于小批量特征交换的三维形状变化自动编码器潜在解纠缠)paper | code  

[3] Interactive Image Synthesis with Panoptic Layout Generation(具有全景布局生成的交互式图像合成) [paper])(https://arxiv.org/abs/2203.02104)  

[2] Polarity Sampling: Quality and Diversity Control of Pre-Trained Generative Networks via Singular Values(极性采样：通过奇异值对预训练生成网络的质量和多样性控制)paper | demo  

[1] Autoregressive Image Generation using Residual Quantization(使用残差量化的自回归图像生成)paper | code  

三维视觉(3D Vision)

[3] The Neurally-Guided Shape Parser: Grammar-based Labeling of 3D Shape Regions with Approximate Inference(神经引导的形状解析器：具有近似推理的 3D 形状区域的基于语法的标记)paper | code  

[2] Deep 3D-to-2D Watermarking: Embedding Messages in 3D Meshes and Extracting Them from 2D Renderings(在 3D 网格中嵌入消息并从 2D 渲染中提取它们)paper  

[1] X -Trans2Cap: Cross-Modal Knowledge Transfer using Transformer for 3D Dense Captioning(使用 Transformer 进行 3D 密集字幕的跨模式知识迁移) 关键词：图像字幕/密集字幕；知识蒸馏；Transformer；三维视觉paper  

点云(Point Cloud)

[10] IDEA-Net: Dynamic 3D Point Cloud Interpolation via Deep Embedding Alignment(通过深度嵌入对齐的动态 3D 点云插值)paper | code  

[9] No Pain, Big Gain: Classify Dynamic Point Cloud Sequences with Static Models by Fitting Feature-level Space-time Surfaces(没有痛苦，收获很大：通过拟合特征级时空表面，用静态模型对动态点云序列进行分类)paper | code  

[8] AutoGPart: Intermediate Supervision Search for Generalizable 3D Part Segmentation(通用 3D 零件分割的中间监督搜索) paper  

[7] Geometric Transformer for Fast and Robust Point Cloud Registration(用于快速和稳健点云配准的几何transformer)paper | code  

[6] Contrastive Boundary Learning for Point Cloud Segmentation(点云分割的对比边界学习)paper | code  

[5] Shape-invariant 3D Adversarial Point Clouds(形状不变的 3D 对抗点云)paper | code  

[4] ART-Point: Improving Rotation Robustness of Point Cloud Classifiers via Adversarial Rotation(通过对抗旋转提高点云分类器的旋转鲁棒性)paper  

[3] Lepard: Learning partial point cloud matching in rigid and deformable scenes(Lepard：在刚性和可变形场景中学习部分点云匹配)paper | code  

[2] A Unified Query-based Paradigm for Point Cloud Understanding(一种基于统一查询的点云理解范式)paper  

[1] CrossPoint: Self-Supervised Cross-Modal Contrastive Learning for 3D Point Cloud Understanding(用于 3D 点云理解的自监督跨模态对比学习) keywords: Self-Supervised Learning, Contrastive Learning, 3D Point Cloud, Representation Learning, Cross-Modal Learningpaper | code  

三维重建(3D Reconstruction)

[10] PLAD: Learning to Infer Shape Programs with Pseudo-Labels and Approximate Distributions(学习用伪标签和近似分布推断形状程序)paper | code  

[9] ϕ-SfT: Shape-from-Template with a Physics-Based Deformation Model(具有基于物理的变形模型的模板形状)paper | code  

[8] Input-level Inductive Biases for 3D Reconstruction(用于 3D 重建的输入级归纳偏差)paper  

[7] AutoSDF: Shape Priors for 3D Completion, Reconstruction and Generation(用于 3D 完成、重建和生成的形状先验)paper | project  

[6] Interacting Attention Graph for Single Image Two-Hand Reconstruction(单幅图像双手重建的交互注意力图)paper | code  

[5] OcclusionFusion: Occlusion-aware Motion Estimation for Real-time Dynamic 3D Reconstruction(实时动态 3D 重建的遮挡感知运动估计)paper | project  

[4] Neural RGB-D Surface Reconstruction(神经 RGB-D 表面重建)paper | project | video  

[3] Neural Face Identification in a 2D Wireframe Projection of a Manifold Object(流形对象的二维线框投影中的神经人脸识别)paper | [code](https://manycore- research.github.io/faceformer) | project  

[2] Generating 3D Bio-Printable Patches Using Wound Segmentation and Reconstruction to Treat Diabetic Foot Ulcers(使用伤口分割和重建生成 3D 生物可打印贴片以治疗糖尿病足溃疡) keywords: semantic segmentation, 3D reconstruction, 3D bio-printerspaper

[1] H4D: Human 4D Modeling by Learning Neural Compositional Representation(通过学习神经组合表示进行人体 4D 建模) keywords: 4D Representation(4D 表征),Human Body Estimation(人体姿态估计),Fine-grained Human Reconstruction(细粒度人体重建)paper

场景重建/视图合成/新视角合成(Novel View Synthesis)

[8] PlaneMVS: 3D Plane Reconstruction from Multi-View Stereo(从多视图立体重建 3D 平面)paper  

[7] NeRFusion: Fusing Radiance Fields for Large-Scale Scene Reconstruction(用于大规模场景重建的融合辐射场)paper  

[6] GeoNeRF: Generalizing NeRF with Geometry Priors(用几何先验概括 NeRF)paper | code  

[5] StyleMesh: Style Transfer for Indoor 3D Scene Reconstructions(室内 3D 场景重建的风格转换)paper | code | project  

[4] Look Outside the Room: Synthesizing A Consistent Long-Term 3D Scene Video from A Single Image(向外看：从单个图像合成一致的长期 3D 场景视频)paper | code | project  

[3] Point-NeRF: Point-based Neural Radiance Fields(基于点的神经辐射场)paper ｜ code |project  

[2] CLIP-NeRF: Text-and-Image Driven Manipulation of Neural Radiance Fields(文本和图像驱动的神经辐射场操作) keywords: NeRF, Image Generation and Manipulation, Language-Image Pre-Training (CLIP)paper | code  

[1] Point-NeRF: Point-based Neural Radiance Fields(基于点的神经辐射场)paper | code | project  

模型压缩(Model Compression)

知识蒸馏(Knowledge Distillation)

[4] Decoupled Knowledge Distillation(解耦知识蒸馏)paper | code  

[3] Wavelet Knowledge Distillation: Towards Efficient Image-to-Image Translation(小波知识蒸馏：迈向高效的图像到图像转换)paper  

[2] Knowledge Distillation as Efficient Pre-training: Faster Convergence, Higher Data-efficiency, and Better Transferability(知识蒸馏作为高效的预训练：更快的收敛、更高的数据效率和更好的可迁移性)paper | code  

[1] Focal and Global Knowledge Distillation for Detectors(探测器的焦点和全局知识蒸馏) keywords: Object Detection, Knowledge Distillationpaper | code  

剪枝(Pruning)

[1] Interspace Pruning: Using Adaptive Filter Representations to Improve Training of Sparse CNNs(空间剪枝：使用自适应滤波器表示来改进稀疏 CNN 的训练)paper  

量化(Quantization)

[2] Implicit Feature Decoupling with Depthwise Quantization(使用深度量化的隐式特征解耦)paper  

[1] IntraQ: Learning Synthetic Images with Intra-Class Heterogeneity for Zero-Shot Network Quantization(学习具有类内异质性的合成图像以进行零样本网络量化)paper | code  

神经网络结构设计(Neural Network Structure Design)

[1] BatchFormer: Learning to Explore Sample Relationships for Robust Representation Learning(学习探索样本关系以进行鲁棒表征学习) keywords: sample relationship, data scarcity learning, Contrastive Self-Supervised Learning, long-tailed recognition, zero-shot learning, domain generalization, self-supervised learningpaper | code  

CNN

[5] TVConv: Efficient Translation Variant Convolution for Layout-aware Visual Processing(用于布局感知视觉处理的高效翻译变体卷积)(动态卷积)paper | code  

[4] On the Integration of Self-Attention and Convolution(自注意力和卷积的整合)paper | code1 | code2  

[3] Scaling Up Your Kernels to 31x31: Revisiting Large Kernel Design in CNNs(将内核扩展到 31x31：重新审视 CNN 中的大型内核设计)paper | code 解读：凭什么 31x31 大小卷积核的耗时可以和 9x9 卷积差不多？ 解读：RepLKNet: 大核卷积+结构重参数让CNN再次伟大  

[2] DeltaCNN: End-to-End CNN Inference of Sparse Frame Differences in Videos(视频中稀疏帧差异的端到端 CNN 推断) keywords: sparse convolutional neural network, video inference acceleratingpaper

[1] A ConvNet for the 2020spaper | code 解读：“文艺复兴” ConvNet卷土重来，压过Transformer！FAIR重新设计纯卷积新架构  

Transformer

[5] Bootstrapping ViTs: Towards Liberating Vision Transformers from Pre-training(引导 ViT：从预训练中解放视觉transformer)paper | code  

[4] Attribute Surrogates Learning and Spectral Tokens Pooling in Transformers for Few-shot Learningpaper | code  

[3] NomMer: Nominate Synergistic Context in Vision Transformer for Visual Recognition(在视觉transformer中为视觉识别指定协同上下文)paper | code  

[2] Delving Deep into the Generalization of Vision Transformers under Distribution Shifts(深入研究分布变化下的视觉Transformer的泛化) keywords: out-of-distribution (OOD) generalization, Vision Transformerspaper | code  

[1] Mobile-Former: Bridging MobileNet and Transformer(连接 MobileNet 和 Transformer) keywords: Light-weight convolutional neural networks(轻量卷积神经网络),Combination of CNN and ViTpaper  

图神经网络(GNN)

神经网络架构搜索(NAS)

[3] Training-free Transformer Architecture Search(免训练transformer架构搜索)paper  

[2] Global Convergence of MAML and Theory-Inspired Neural Architecture Search for Few-Shot Learning(MAML 的全局收敛和受理论启发的神经架构搜索以进行 Few-Shot 学习)paper | code  

[1] β-DARTS: Beta-Decay Regularization for Differentiable Architecture Search(可微架构搜索的 Beta-Decay 正则化)paper  

MLP

[3] Dynamic MLP for Fine-Grained Image Classification by Leveraging Geographical and Temporal Information(利用地理和时间信息进行细粒度图像分类的动态 MLP)paper | code  

[2] Revisiting the Transferability of Supervised Pretraining: an MLP Perspective(重新审视监督预训练的可迁移性：MLP 视角)paper  

[1] An Image Patch is a Wave: Quantum Inspired Vision MLP(图像补丁是波浪：量子启发的视觉 MLP)paper | code | code  

数据处理(Data Processing)

[1] Dataset Distillation by Matching Training Trajectories(通过匹配训练轨迹进行数据集蒸馏)(数据集蒸馏)paper | code | project  

数据增广(Data Augmentation)

[2] TeachAugment: Data Augmentation Optimization Using Teacher Knowledge(使用教师知识进行数据增强优化)paper ｜ code  

[1] 3D Common Corruptions and Data Augmentation(3D 常见损坏和数据增强) keywords: Data Augmentation, Image restoration, Photorealistic image synthesispaper | projecr  

归一化/正则化(Batch Normalization)

[1] Delving into the Estimation Shift of Batch Normalization in a Network(深入研究网络中批量标准化的估计偏移)paper | code  

图像聚类(Image Clustering)

[1] RAMA: A Rapid Multicut Algorithm on GPU(GPU 上的快速多切算法)paper | code  

图像压缩(Image Compression)

[4] Unified Multivariate Gaussian Mixture for Efficient Neural Image Compression(用于高效神经图像压缩的统一多元高斯混合)paper | code  

[3] ELIC: Efficient Learned Image Compression with Unevenly Grouped Space-Channel Contextual Adaptive Coding(具有不均匀分组的空间通道上下文自适应编码的高效学习图像压缩)paper  

[2] The Devil Is in the Details: Window-based Attention for Image Compression(细节中的魔鬼：图像压缩的基于窗口的注意力)paper | code  

[1] Neural Data-Dependent Transform for Learned Image Compression(用于学习图像压缩的神经数据相关变换)paper | code | project  

模型训练/泛化(Model Training/Generalization)

[6] Out-of-distribution Generalization with Causal Invariant Transformations(具有因果不变变换的分布外泛化)paper  

[5] Can Neural Nets Learn the Same Model Twice? Investigating Reproducibility and Double Descent from the Decision Boundary Perspective(神经网络可以两次学习相同的模型吗？ 从决策边界的角度研究可重复性和双重下降)paper | code  

[4] Towards Efficient and Scalable Sharpness-Aware Minimization(迈向高效和可扩展的锐度感知最小化) keywords: Sharp Local Minima, Large-Batch Trainingpaper  

[3] CAFE: Learning to Condense Dataset by Aligning Features(通过对齐特征学习压缩数据集) keywords: dataset condensation, coreset selection, generative modelspaper | code  

[2] The Devil is in the Margin: Margin-based Label Smoothing for Network Calibration(魔鬼在边缘：用于网络校准的基于边缘的标签平滑)paper | code  

[1] DN-DETR: Accelerate DETR Training by Introducing Query DeNoising(通过引入查询去噪加速 DETR 训练) keywords: Detection Transformerpaper | code  

噪声标签(Noisy Label)

[2] Scalable Penalized Regression for Noise Detection in Learning with Noisy Labels(带有噪声标签的学习中噪声检测的可扩展惩罚回归)paper | code  

[1] Scalable Penalized Regression for Noise Detection in Learning with Noisy Labels(Scalable Penalized Regression for Noise Detection in Learning with Noisy Labels)paper | code  

长尾分布(Long-Tailed Distribution)

[1] Targeted Supervised Contrastive Learning for Long-Tailed Recognition(用于长尾识别的有针对性的监督对比学习) keywords: Long-Tailed Recognition(长尾识别), Contrastive Learning(对比学习)paper  

图像特征提取与匹配(Image feature extraction and matching)

[1] Probabilistic Warp Consistency for Weakly-Supervised Semantic Correspondences(弱监督语义对应的概率扭曲一致性)paper | code  

视觉表征学习(Visual Representation Learning)

[4] Node Representation Learning in Graph via Node-to-Neighbourhood Mutual Information Maximization(通过节点到邻域互信息最大化的图中节点表示学习)paper | code  

[3] SimAN: Exploring Self-Supervised Representation Learning of Scene Text via Similarity-Aware Normalization(通过相似性感知归一化探索场景文本的自监督表示学习)paper  

[2] Exploring Set Similarity for Dense Self-supervised Representation Learning(探索密集自监督表示学习的集合相似性)paper  

[1] Motion-aware Contrastive Video Representation Learning via Foreground-background Merging(通过前景-背景合并的运动感知对比视频表示学习)paper | code  

模型评估(Model Evaluation)

多模态学习(Multi-Modal Learning)

[1] MERLOT Reserve: Neural Script Knowledge through Vision and Language and Sound(通过视觉、语言和声音的神经脚本知识)paper | project  

视听学习(Audio-visual Learning)

视觉-语言（Vision-language）

[8] An Empirical Study of Training End-to-End Vision-and-Language Transformers(培训端到端视觉和语言transformer的实证研究)paper | code  

[7] Pseudo-Q: Generating Pseudo Language Queries for Visual Grounding(为视觉基础生成伪语言查询)paper | code  

[6] Conditional Prompt Learning for Vision-Language Models(视觉语言模型的条件提示学习)paper | code  

[5] NLX-GPT: A Model for Natural Language Explanations in Vision and Vision-Language Tasks(视觉和视觉语言任务中的自然语言解释模型)paper | code  

[4] L-Verse: Bidirectional Generation Between Image and Text(图像和文本之间的双向生成) (Oral Presentation)paper  

[3] HairCLIP: Design Your Hair by Text and Reference Image(通过文本和参考图像设计你的头发) keywords: Language-Image Pre-Training (CLIP), Generative Adversarial Networkspaper | project  

[2] CLIP-NeRF: Text-and-Image Driven Manipulation of Neural Radiance Fields(文本和图像驱动的神经辐射场操作) keywords: NeRF, Image Generation and Manipulation, Language-Image Pre-Training (CLIP)paper | code  

[1] Vision-Language Pre-Training with Triple Contrastive Learning(三重对比学习的视觉语言预训练) keywords: Vision-language representation learning, Contrastive Learning paper | code  

视觉预测(Vision-based Prediction)

[7] Remember Intentions: Retrospective-Memory-based Trajectory Prediction(记住意图：基于回顾性记忆的轨迹预测)paper | code  

[6] GaTector: A Unified Framework for Gaze Object Prediction(凝视对象预测的统一框架)paper  

[5] On Adversarial Robustness of Trajectory Prediction for Autonomous Vehicles(自动驾驶汽车轨迹预测的对抗鲁棒性)paper | code  

[4] Adaptive Trajectory Prediction via Transferable GNN(基于可迁移 GNN 的自适应轨迹预测)paper  

[3] Towards Robust and Adaptive Motion Forecasting: A Causal Representation Perspective(迈向稳健和自适应运动预测：因果表示视角)paper | code  

[2] How many Observations are Enough? Knowledge Distillation for Trajectory Forecasting(多少个观察就足够了？ 轨迹预测的知识蒸馏) keywords: Knowledge Distillation, trajectory forecastingpaper  

[1] Motron: Multimodal Probabilistic Human Motion Forecasting(多模式概率人体运动预测)paper  

数据集(Dataset)

[6] M5Product: Self-harmonized Contrastive Learning for E-commercial Multi-modal Pretraining(电子商务多模态预训练的自协调对比学习)(多模态预训练数据集)paper  

[5] FERV39k: A Large-Scale Multi-Scene Dataset for Facial Expression Recognition in Videos(用于视频中面部表情识别的大规模多场景数据集)paper  

[4] Ego4D: Around the World in 3,000 Hours of Egocentric Video(3000 小时以自我为中心的视频环游世界)paper | project  

[3] GrainSpace: A Large-scale Dataset for Fine-grained and Domain-adaptive Recognition of Cereal Grains(用于细粒度和域自适应识别谷物的大规模数据集)paper | dataset  

[2] Kubric: A scalable dataset generator(Kubric：可扩展的数据集生成器)paper | co