2021计算机视觉-包揽所有前沿论文源码 -上半年

大家是否遇到过这种情况，就是在工作或者学习的时候，想去找一些方向的网络，但是呢，尴尬的是，老旧的网络里不想要，前沿的网络又不知道有哪些。为了解决大家的这个困扰，本人决定收集2021年上半年大部分前沿的网络相关链接，之后我会( $\color{blue}{文末附带}$ $\color{blue}{公众号 -}$ $\color{blue}{ 海量资源}。$ )：

$\color{red}{每周一更新一次（下面是我多年年收集的链接地址） }$
计算机视觉-包揽所有前沿论文源码

有兴趣的朋友可以加微信：17575010159 相互讨论技术。若是帮助到了你什么，一定要记得点赞！因为这是对我最大的鼓励!

视觉工作项目-为后来的你，提供一份帮助！
$\color{red}{上面这个链接是我所有工作项目的详细解。}$

文章分类

（01）AAAI 2021 | 腾讯优图11篇论文入选，涵盖动作识别、人群密度估计、人脸安全等领域
（02）重磅！网易伏羲9篇论文入选AI顶会AAAI 2021
（03）CVPR2020 最全整理：论文汇总 / 代码 / 项目 / 论文解读（更新中）【计算机视觉】
（04）CVPR、ECCV 2020 两大会议论文分类索引
（05）人体姿态估计、识别与生成最新技术一览
（06）一文概览 CVPR2021 最新18篇 Oral 论文
（07）WACV 2021 论文大盘点-GAN 篇与行人监控篇
（08）近期必看的视觉综述，含 GAN、Transformer、人脸超分辨、遥感等
（09）CVPR2021 最全整理：论文汇总 / 代码 / 项目 / 论文解读（更新中）【计算机视觉】
（10）重磅！悉尼科大ReLER实验室13篇论文入选CVPR 2021
（21）WACV 2021 论文大盘点目标检测与图像分割篇（持续更新）
（22）WACV 2021 论文大盘点-GAN 篇与行人监控篇
（23）近期必看的视觉综述，含图像检索、目标检测、人脸关键点检测、医学图像分割、遥感、模型优化等
（24）WACV 2021 Paper Inventory - Human Action Detection and Recognition & Image and Video Retrieval
(25) AAAI 2021 | Summary of excellent papers from Microsoft Research Asia!
(26) An overview of the latest 18 Oral papers of CVPR2021
(27) The most comprehensive arrangement of CVPR2021: paper summary / code / project / paper interpretation (updating) [computer vision]
(28) CVPR2021 latest collection of papers received! Summary of 100+ papers in 22 directions｜Continuous update
(29) Recommend several must-read visual reviews in the near future, including GAN, Transformer , face super-resolution, remote sensing, etc. (
30) Recommend several recent must-read visual reviews, including image retrieval, target detection, face key point detection, medical image segmentation, remote sensing, model optimization, etc. ! Continuously updating! (33) https://github.com/52CV/CVPR-2021-Papers

（34）CVPR2021中的目标检测和语义分割论文汇总
（35）一文概览 CVPR2021 最新18篇 Oral 论文
（36）CVPR 2021 | 腾讯AI Lab入选论文解读
（37）顶会论文分类汇总，包含WACV21/CVPR19、20/ECCV20（附下载）
（38）2021 最新CV综述分类汇总（持续更新）
（39）CVPR 2021 论文/代码分类汇总！持续更新中！
（40）CVPR 2021 速览 | 旷视研究院22篇入选学术成果盘点
（41）一文概览 CVPR2021 最新18篇 Oral 论文
（42）CVPR 2021放榜，腾讯优图20篇论文都在这里了！
（43）CVPR 二十年，影响力最大的10篇论文！
（44）CVPR 2021公布最佳论文候选！华人占据半壁江山，何恺明、沈春华等人上榜
（45）添加链接描述CVPR 2021大奖出炉！何恺明获最佳论文提名，华人四篇“最佳”！第一届Thomas S. Huang 纪念奖颁发）
（46）CVPR 二十年，影响力最大的10篇论文！
（47）Just now, CVPR 2021 Best Paper, Best Student Paper and other awards have been announced! (Paper download address is attached)
(48) CVPR 2021 awards released: Best paper goes to Max Planck Institute, He Yuming was nominated, and the first Huang Xutao Memorial Award was announced
(49) Recommended open source papers this week: including face recognition, instance segmentation, tracking, SR, etc.
(50) CVPR 2021 papers are open for download!
(51) [CVPR 2021 Best Paper Candidate] 32 best paper candidates have been announced. Guess which one won the CVPR 2021 Best Paper Candidate?
(52) [June 2] Code sharing of ten (to be) open source papers
(53) May 26] Code sharing of seven (to be) open source papers
(56) 2021 745 published papers are the most comprehensive classification summary!
(57) Recommend several new CVPR 2021 open source papers, including image segmentation, domain adaptation, image retrieval, line of sight estimation, etc.

face technology

(01) The best new framework of CVPR2020｜Large-scale facial expression recognition (with source code)
(02) Inventory｜Lightweight face detection algorithm implementation, almost no friends are here~
(03) 10 lightweight face detection algorithms big PK｜Code open source
(04) The remaining problems of face recognition: from occlusion, age, posture, makeup to kinship, face attack
(05) Overview of liveness detection algorithms in face recognition
(0) 6) TinaFace: A new record for face detection!
(07) The accuracy rate exceeds 99.5%!

如何入门多视角人脸正面化生成？不得不看的超详细最新综述！
（07）WACV 2021 论文大盘点-人脸技术篇
（08）重要！分享几个业界新出人脸识别数据集
（09）Facebook等新提出的视听语音分离的方法VisualVoice，利用跨模态一致性
VisualVoice: Audio-Visual Speech Separation with Cross-Modal Consistency
单位 |德克萨斯大学奥斯汀分校，Facebook
论文 |https://arxiv.org/abs/2101.03149
代码 |https://github.com/facebookresearch/VisualVoice
主页 |http://vision.cs.utexas.edu/projects/VisualVoice/
（10）人脸超分辨率，基于迭代合作的方法
（11）不得不赞！京东开源FaceX-Zoo，一站式人脸识别研究平台
（12）跳过人脸检测和关键点定位，Facebook等提出实时3D人脸姿态估计新方法
（13）无需人脸检测和关键点定位，Facebook等提出实时3D人脸姿态估计新方法
（14）CVPR 2021 | 中科大联合快手，提出人脸伪造检测新方法
（15）Face Transformer for Recognition is used for face recognition
(16) The Chinese team won the world's first face recognition mask!
(17) Open source! Face detection algorithm with only 85K parameters
(18) Occlusion face problem | Detailed interpretation of Attention-Based method to solve occlusion face recognition problem (attached paper download)
(19) CVPR2021 (Oral) Shang Tang, Hong Kong Chinese to achieve a new breakthrough in monocular face reconstruction: a renderer based on a generative network! Geometry is more precise! The rendering effect is more realistic!
(20) OpenVINO™ realizes eye fatigue/drowsiness detection based on facial landmark detection
(21) Tencent Youtu TFace is officially open source, more reliable face recognition!
(22) CVPR2021 (Oral) Shangtang and Hong Kong Chinese achieve a new breakthrough in monocular face reconstruction: a renderer based on a generative network! Geometry is more precise! The rendering effect is more realistic!
(23) Monocular 3D face reconstruction, wrinkles can change naturally with the expression, more realistic.
(24) 3D face modeling Snap et al. proposed the first one-shot 3D face style migration framework, which only needs an image of any style, and it can generate a 3D face model with exaggerated geometric shape and texture stylization.

Target Detection

(01) Video target detection inventory
(02) R-CenterNet: Use CenterNet to detect rotating targets
(03) The University of Hong Kong proposed OneNet: a one-stage end-to-end target detection network without NMS! No binary matching required!
(04) Overview of Anomaly Detection
(05) Open Source Software | Deep Learning for Road Disease Detection
(06) Transformer has made another contribution! Fast (420 fps) and good lane line detection algorithm
(07) NanoDet: Lightweight (1.8MB), ultra-fast (97fps on mobile) object detection project
(08) Use your strength to justify yourself, YOLOv5: I am the strongest in road damage detection! GRDDC'2020 Contest Report
(09) An alternative to YOLO, the 97FPS Anchor-Free target detection model NanoDet on the mobile phone is now open source~
(10) Excellent! Tongji Berkeley of Hong Kong University proposed Sparse R-CNN: a new paradigm for target detection
(11) Generalized Focal Loss V2 for painless target detection
(12) Using CenterNet to detect rotating targets
(13) Crack detection scheme based on computer vision
(14) Multimedia Laboratory of Chinese University of Hong Kong | Open source video target detection & tracking platform (with source code download)
（15）基于密度图的航空物体检测：理论与代码实现
（16）目标检测的稀疏对抗攻击，代码已开源
（17）北亚利桑那大学等推出：航拍森林火情检测数据集 FLAME
（18）无需NMS的目标检测，OneNet
（19）NAS在检测中的应用
（20）NeurIPS 2020 | 微软亚洲研究院论文摘录之目标检测篇
（21）难以置信的目标检测小妙招：多训练几个epochs，平均一下就能获得更好的模型

(22) Sparse adversarial attack for object detection, the code has been open-sourced
(23) C++ implements OpenVINO deployment of yolov5
(24) Jishi live broadcast playback 丨 No. 75 - Fang Hao: New SOTA for lane line detection, RESA: cyclic feature displacement aggregator (AAAI2021) (25) One article sorts out defect detection methods (26) Open source project | Based on YOLO-V5 to realize pedestrian social distance risk warning (with complete source code
) (
27 )
Heavyweight ! 13 Anchor free-based target detection methods
(28) Interpretation of rotating target detection methods (DCL, CVPR2021)
(29) One article sorting out defect detection methods
(30) No NMS! Alibaba and Ada proposed PSS: a simpler and more effective end-to-end target detection
(14) point-up technique! Small target detection: data augmentation
(15) AAAI 2021 target detection paper inventory (YOLObile/R3Det/StarNet, etc.)
(16) Target detection competition ideas, tricks collection, data summary
(17) CVPR 2021 | GFLV2: Target detection conscience technology, no Cost increase point!
(18) Detailed practical tutorial: Deploying YOLOv5 target detection with OpenCV's DNN module
(19)Dry goods practice | Anchor optimization improves target detection so significantly
(20) Small object problem in object detection
(21) Big change to Yolo framework | New target detection framework with extremely low energy consumption (attached paper download) (
22) Small target detection: data enhancement
(23) Big inventory | 2 best reviews of abnormal algorithms in 2020 (
24) Small target detection: Feature Extraction
(25) Summary of the latest research on industrial image anomaly detection (2019-20) 20)
(26) Overview丨Research progress on surface defect detection of industrial metal flat materials
(27) Detailed explanation of camouflaged target detection based on deep learning
(28) Deployment of YOLOV5 model based on Caffe format
(29) #WACV 2021 FisheyeYOLO: Universal Object Detection on Fisheye Cameras for Autonomous Driving. For object detection in fisheye images, the author found better representation methods in different object representation methods, such as oriented bounding boxes, ellipses and general polygons. And a novel curved bounding box model is designed, which has the best properties of the fisheye distortion model. FisheyeYOLO: Generalized Object Detection on Fisheye Cameras for Autonomous Driving unit | University of Limerick, Valeo paper | https://www.researchgate.net/publication/346931586_FisheyeYOLO_Object_Detection_on_Fisheye_Cameras_for_Autonomous_Driving code |

(30) Aerial object detection based on density map: theory and code implementation
(31) WACV 2021 paper inventory - target detection
(32) Target detection in AAAI 2021 (detailed version with code)
(33) From L1 loss to EIoU loss, a list of loss functions for target detection border regression
(34) #城市天眼# Developed by Skylark Labs in the United States, drone security monitoring can be used in high altitude (3- 90 meters) to detect and analyze behaviors of crowds and find suspicious activities.
Source: https://twitter.com/i/status/1364086835266211843
(35) Fast and accurate without lidar! SMOKE for 3D target detection
(36) Understanding Objectness in object detection

（37）目标检测一卷到底之后，终于有人为它挖了个新坑｜CVPR2021 Oral
（38）CVPR2021目标检测佳作 | Weighted boxes fusion（附github源码及论文下载）
（39）基于YOLOV4深度网络的车辆压实线检测算法
（40）56.4 AP！超越YOLOv4，更快更强的CenterNet2来了！
（41）CVPR2121目标检测 | 少见的知识蒸馏用于目标检测（附论文下载）
（42）用于自动驾驶的实时车道线检测和智能告警
（43）全新FPN！CE-FPN：通道增强特征金字塔网络，助力目标检测涨点！
（44）极市项目｜未拴绳遛狗识别算法需求
（45）基于YOLOV5深度网络模型的火焰检测
（46）基于YOLOV5深度网络模型的交通标志设施的模型训练
（47）基于深度学习YOLOV5网络的道路状况检测
（48）基于YOLOV5深度网络的公路病害检测
（49）使用Disentangling形式的损失函数回归2D和3D目标框
（50）CVPR 2021 | 腾讯AI Lab入选论文解读
（51）实操教程：android camera nanodet 实时物体检测的高效实现总结
（52）CVPR2021 目标检测佳作 | Weighted boxes fusion（附 GitHub 源码及论文下载）
（53）我扔掉FPN来做目标检测，效果竟然这么强！YOLOF开源：你只需要看一层特征｜CVPR2021
（54）【入门教程】异常检测（Anomaly Detection）到底是什么？
（55）最强检测 | YOLO V4?都是弟弟! CenterNet2以56.4mAP超越当前所有检测模型
（56）mmdetection性能简单优化方法
（57）目标检测一卷到底之后，终于有人为它挖了个新坑｜CVPR2021 Oral
（58）轻量高速检测器LFFD升级版LFD发布！用Pytorch部署，支持多类检测
（59）船舶检测 | 计算机视觉来看苏伊士运河堵船
（60）基于YOLOV4的印刷电路板PCB目标检测
（61）INT4量化用于目标检测
（62）超越YOLOv5！PP-YOLOv2：更快更好的目标检测网络
（62）Hugging Face发布PyTorch新库「Accelerate」：适用于多GPU、TPU、混合精度训练
（63）超越YOLOv5还不够！这个目标检测开源项目又上新了
（64）60.6 AP！打破COCO记录！微软提出DyHead：将注意力与目标检测Heads统一
（65）当YOLOv5遇见OpenVINO！
（66）OpenVINO™ 头部姿态评估网络应用演示
（67）实操教程｜YOLOv5实现自定义对象训练与OpenVINO部署全解析
（68）缺陷检测算法汇总（传统+深度学习方式）|综述、源码
（69）一文梳理水下目标检测方法
（70）不容忽视的问题：行人检测器的泛化能力
（71）让检测告别遮挡 | NMS-Loss是如何解决目标检测中的遮挡问题的？
（72）旋转目标检测 | 基于高斯 Wasserstein 距离损失的目标检测（附源代码）
（73）干货 | 利用像机图像通过卷积神经网络实时进行水稻检测（致敬袁老）
（74）MaskedFace-Net | 新冠疫情中的口罩检测（附论文及源代码）
（75）CVPR 2021 | 谷歌提出MobileDets：轻量化目标检测网络
（76）收藏 | 使用合成数据集做目标检测
（77）Moving target detection - ViBe algorithm
(78) Selected series of target detection, the most complete summary at present!

Classification, re-identification (backbone network)

(01) NanoDet, a 1.8M ultra-light object detection model, runs faster than YOLO, and has more than 200 Stars in two days. (
02) Sun Yat-sen University proposes a new pedestrian re-identification method and the largest evaluation benchmark in history
(03) ECCV 2020 paper inventory - remote sensing and aerial image processing and recognition
(04) Sun Yat-sen University proposes a new pedestrian re-identification method and the largest evaluation benchmark in history
(05) Video person re-identification: relationship guides spatial attention + temporal features Extraction model
(06) Wuhan University and others released the latest review of ReID! Including the three top visual conferences, a new benchmark method AGW｜TPAMI2021
(07) Sun Yat-sen University proposed a new pedestrian re-identification method and the largest evaluation benchmark in history
(08) was fully upgraded! FastReID V1.0 is officially open source: Beyond reID
(09) the strongest ResNet variant! Goodbye normalization! DeepMind proposed NFNet, the code has been open source!
(10) Review and Prospect of Deep Learning Pedestrian Re-Identification, Latest Articles of TPAMI 2021
(11) Supervised Pedestrian Re-Identification Problem in Camera Domain
(12) WACV 2021 Paper Review - Image Classification
(13) WACV 2021 Paper Review - Image and Video Retrieval
(14) CVPR 2021 | Target-Guided Human Attention Estimation Improves Zero-Shot Learning
(15)From the road to simplicity! In-depth interpretation of CVPR2021 paper RepVGG!
(16) Propose an end-to-end prototypical cross-domain self-supervised learning (PCS) framework for Few-shot Unsupervised Domain Adaptation (FUDA).
(17) ResNet has been strongly upgraded, and it will compete with EfficientNets only by improving training and expansion strategies

(19) CVPR2021|ACNet evolved again, Tsinghua University & Megvii Technology proposed an Inception type DBB
(20) After two years, EfficientNet v2 is here! Faster, smaller and stronger!
(21) 89.77% accuracy rate! Google proposed CoAtNet: Combining convolution and self-attention
(22) CVPR 2021 Oral | A new self-attention model beyond convolution! Google proposes: HaloNet, another super-strong visual backbone...
(23) Research status and problems of remote sensing image classification of hyperspectral images
(24) Google proposes a new model of "convolution + attention", surpassing the strongest variant of ResNet!
(25) The world's first open source image recognition system is launched
(26) Bytedance won the double champion of CVPR2021 fine-grained image competition
(27) Dry goods | Ali's image search architecture
(28) EfficientNetV2

Semantic Object Segmentation

(01) NeurIPS 2020 Oral: Using Pixel-Level Cyclic Consistency to Solve Domain Adaptive Semantic Segmentation Problems
(02) Performance Improvement by More Than 30%! Industrial SOTA's real-time instance segmentation algorithm SOLOv2, faster and stronger!
(03) Inventory of CVPR 2020 papers - matting
(04) Real-time matting without green screen, SenseTime and others proposed a new method MODNet that only needs a single image and a single model (
05) performance improvement of more than 30%, real-time instance segmentation algorithm SOLOv2 realizes industry SOTA
(06) YolactEdge, the first real-time instance segmentation method on edge devices (Jetson AGX Xavier: 30 FPS
(07) medical image segmentation Comprehensive comparison of the best methods: U-Net and U-Net++
(06) MODNet is easy to train in an end-to-end fashion. It is much faster than contemporaneous matting methods, running at 63 frames per second.

(07) In this work, the author proposed BoxInst, which can only be labeled with instance bounding boxes (rather than instance mask labels)
(08) The author named this joint task as depth-aware video panorama segmentation, and proposed a new evaluation index and two derived datasets for it, and said that these datasets will be made public.

Transformer再突破！MedT：医学图像分割新网络
（22）CVPR 2021 | MSRA提出像素级别自监督预训练方法PixPro，大幅提升下游检测分割任务性能
（23）SG-net：一次视频实例分割的空间粒度网络
（24）Panoptic FCN：真正End-to-End的全景分割
（25）CVPR 2021 Oral | Transformer再突破！美团等提出VisTR：视频实例分割网络
（26）CVPR 2021 | 250 FPS！让实时语义分割飞起！重新思考BiSeNet
（27）顶刊TPAMI 2021！南开大学提出深度霍夫变换：语义线检测新方法
（28）Segmenter：基于纯Transformer的语义分割网络
（29）谷歌等新作：视觉Transformer的有趣特性
（30）视觉Transformer比CNN更鲁棒！IBM华人研究员新作
（31）更快更强！谷歌提出NesT：收敛更快、鲁棒更好的Transformer

(32) real-time, high-resolution background replacement techniques that run at 30fps in 4K resolution and 60fps in HD, and the code is open source!
(33) HKU & NVIDIA proposed SegFormer: a new idea for simple and effective Transformer semantic segmentation
(34) Practical tutorial｜An example of using image segmentation for defect detection
(35) CVPR2021 masterpiece | One-Shot is too much, Zero-Shot instance sample segmentation
(36) Training data does not need to be manually labeled and segmented, but can image segmentation also be achieved?
(37) CVPR2021 double-layer instance segmentation, greatly improving occlusion processing performance
(38) Google releases a new dataset for semantic segmentation! By the way, a model slaughter list was developed, which has been accepted by CVPR2021

target tracking

(01) Favorites | Introduction to Multi-Target Tracking (MOT)
(02) Overview of Single-Target Tracking
(03) Simple and crude Multi-Target Tracking Artifact – DeepSort
(04) Long-term target tracking combined with re-detection
(05) Target tracking has added a heavy open source toolbox, MMTracking is here!
(06) Remote sensing image + CNN, predicting regional population income level
(07) Summary of target tracking
(08) WACV 2021 paper inventory - target tracking

(09) Inadvertently "walking two steps" can lock identity information, which is the black technology of gait recognition.
Recommend a new review that comprehensively introduces the development of gait recognition, including technology evolution, main data sets, and the current level of technology. It is a must-read paper for understanding deep learning gait recognition.
Deep Gait Recognition: A Survey https://arxiv.org/pdf/2102.09546.pdf

(10) #手身追输# Human hands are extremely flexible, and there are various complex self-contacts and occlusions, which bring difficulties to tracking. Facebook Reality Labs recently invented an extremely accurate method of hand tracking by adding physical constraints to the visual model. Highly accurate tracking is possible with one or two hands.
Constraining Dense Hand Surface Tracking with Elasticity
Home | https://research.fb.com/publications/constraining-dense-hand-surface-tracking-with-elasticity/

(11) TraDeS: CVPR 2021 multi-target tracking algorithm, which improves the current online method of joint detection and tracking, uses tracking clues to assist detection, and achieves a significant improvement in accuracy in multiple data sets. The author is from the State University of New York. The paper has not yet been published, and the code will be open source.
Track to Detect and Segment: An Online Multi-Object Tracker
project homepage: https://jialianwu.com/projects/TraDeS.html

(12) Multi-channel surveillance video stitching system based on scale-invariant feature transformation
(13) TCSVT2021: A pedestrian re-identification method combining global and local fine-grained features
(14) Image stitching algorithm based on SIFT scale-invariant feature transformation
(15) The latest open source! TransReID: The first ReID network based on Transformer, leading in all tasks!
(16) Interpretation of WACV2021 paper - Scale Equivariance Improves Siamese Tracking
(17) End-to-end multi-object tracking, the code will be open source, Looking Beyond Two Frames: End-to-End Multi-Object Tracking Using Spatial and Temporal Transformers (18) SiamGAT is proposed for
object tracking, and its performance is ahead of many current advanced trackers, reaching SOTA .
(19) TCSVT2021: A pedestrian re-identification method that combines global and local fine-grained features
(20) CVPR 2021 | The first pedestrian search framework without an anchor frame (Anchor-Free) (with code)
(21) Pedestrian multi-target tracking based on YOLOV3 and DeepSort
(22) From theory to actual combat! Video Streaming Vehicle Counting and Object Tracking
(23)TPAMI 2021 ：基于 event stream 的步态识别，准确率高达90%
（24）极市直播回放丨第80期-张新宇：CVPR 2021-Alpha Refine：通过精确的边界框估计提高跟踪性能
（25）目标跟踪入门篇-相关滤波

动作检测与识别

（01）MMAction2: 新一代视频理解工具箱
（02）WACV 2021 论文大盘点-人体动作检测与识别篇
（03）CVPR 2021 | 用于动作识别，即插即用、混合注意力机制的 ACTION 模块
（04）CVPR 2021 | 商汤提出最强时序动作提名修正网络：TCANet
（05）人体动作识别与生成：基于ST-GCN的方法
（06）刷爆HACS挑战赛时序动作检测榜单！TCANet：最强时序动作提名修正网络 CVPR 2021
（07）更快更强！视频理解模型PP-TSM重磅发布：速度比SlowFast快4.5倍
（08）视频异常行为检测算法MPN，在多个数据库上达到SOTA
（09）CVPR2021Oral #人体运动捕捉使用 4 个RGBD摄像头进行人体运动捕捉，在几何重建和纹理生成上效果都更好
（10）CVPR 2021 | 又好又快的视频异常检测，引入元学习的动态原型学习组件

姿态估算

（01）多人姿态识别框架——AlphaPose
（02）GitHub：人体姿态估计最全资料集锦
（03）人体姿态估计 (Human Pose Estimation) 常用方法总结
（04）CVPR2020 | 旷视研究院提出 PVN3D：基于 3D 关键点投票网络的单目 6DoF 位姿估计算法
（05）人体姿态估计、识别与生成最新技术一览
（06）深度学习人体姿态估计：2014-2020全面调研
（07）最新开源：端到端6D物体姿态跟踪，无需标注数据集！
（08）手势识别基础～手势骨架与关键点提取
（09）动物姿态估计！马、老虎、牛、鹿、狗狗的姿态都能搞定！斩获CVPR 2021 Oral
（10）OpenVINO™ 头部姿态评估网络应用演示
（11）CVPR 2021 | 微软提出"解构式关键点回归"，刷新COCO自底向上多人姿态检测记录！

OCR

（01）万字长文 | 图表示学习中的Encoder-Decoder框架
（02）霸榜Github：又一款OCR神器面世！
（03）新视角：用图像分类来建模文字识别也可以SOTA
（05）都2021了，别再堆砌网络了！10万奖金悬赏最强轻量化OCR模型
（06）顶刊TPAMI 2021！PAN++：精确高效的任意形状文本检测与识别
（07）最新！CVPR 2021 OCR领域论文大盘点（22篇）
（08）论文推荐｜【KSII TIIS 2021】DP-LinkNet：一种用于古籍文档图像二值化的卷积网络(有源码)

3D,深度估算,点云,SLAM

（01）CVPR2020 | 3D 目标检测新框架：3DSSD
（02）CenterFusion：融合雷达与摄像头数据的高精度3D目标检测
（03）最佳论文！商汤提出手机端实时单目三维重建系统 | ISMAR 2020

（04）商汤提出手机端实时单目三维重建系统，实现逼真AR效果和交互
（05）基于深度学习的图像匹配技术一览
（06）极市直播｜AAAI’21杰出论文许鸿斌：一个解决三维重建对数据依赖的新框架（已开源）
（07）OpenCV再升级！修改一行代码，将图像匹配效果提升14%！
（08）重磅！谷歌开源TensorFlow 3D场景理解库
（09）极市直播回放丨第76期-许鸿斌：AAAI’21杰出论文，一个解决三维重建对数据依赖的新框架（已开源）
（10）可用于大规模点云表面重建的深度学习算法
（11）可用于大规模点云表面重建的深度学习算法

（12）深度估计是机器人和自动驾驶研究的重要内容，而这往往需要特殊设备，如RGB-D相机或激光雷达，如何使用RGB相机感知深度呢？研究人员曾经做了很多的尝试。该视频是CVPR 2021论文Depth from Camera Motion and Object Detection结果，通过使用“普通手机摄像头运动+目标检测的包围框”数据，设计RNN网络实现了达到最先进精度的目标深度估计。单位 | 密歇根大学，史蒂文森理工学院论文 | https://arxiv.org/abs/2103.01468代码 | https://github.com/griffbr/ODMD
（13）CVPR 2021 | TPCN 点云就是这么美妙
（14）一文了解激光点云的组织形式
（15）基于YOLO的新型RGB-D融合方法对行人进行检测和3D定位
（16）ECCV2020 | 夜间图像的无监督单目深度估计
（17）MVSNeRF: Fast Generalizable Radiance Field Reconstruction from Multi-View Stereo
（18）在 KITTI 基准数据集上实现最先进的单目3D目标检测结果，表现与基于单目视频的方法相当。
（19）真正实用的退化模型：ETH开源业内首个广义盲图像超分退化模型，性能效果绝佳
（20）ResNet也能用在3D模型上了！清华计图首创三角网格面片上的卷积神经网络：SubdivNet
（21）开源｜AAAI‘21杰出论文-三维重建新探索：解决数据依赖问题，让自监督信号更可靠！
（22）综述：基于点云的自动驾驶3D目标检测和分类方法
（23）PatchmatchNet：一种高效multi-view stereo框架 (CVPR2021 Oral)
（24）CVPR2021｜神经网络如何进行深度估计？
（25）DXSLAM：一种基于深度特征的鲁棒且高效的视觉SLAM系统
（26）实时高分辨率 RGB-D表面重建（CVPR2021）
（27）Complexer-YOLO：基于语义点云的实时三维目标检测与跟踪
（28）HDRUNet | 深圳先进院董超团队提出带降噪与反量化功能的单帧HDR重建算法
（29）基于点云的3D障碍物检测
（30）极市直播丨朱思语：基于深度学习的视觉稠密建图和定位
（31）基于3D Surfel图的单目直接法稀疏定位
（32）将合成 3D 场景表示合并到生成模型中，从而实现更可控的图像合成。
（33）传统单图像深度估计往往只能给出低分辨率结果，细节也不够丰富，视觉上总给人模糊不清的感觉，来自SFU和Adobe的研究者通过合并不同分辨率生成高分辨率的深度估计，终于可以还原清晰的细节。
(34) OmniPhotos, currently the fastest 360° panoramic VR photography method. The code is open source.
(35) The author proposes and integrates GrooMeD-NMS – a novel grouped mathematically differentiable NMS for monocular 3D object detection, (
36) CVPR 2021 | Adaptive Activation Function ACON: A New Paradigm for Unifying ReLU and Swish

GUN (image generation, super resolution, motion transfer)

(01) CVPR 2020 paper inventory - image enhancement and image restoration
(02) Harbin Institute of Technology and others proposed a lightweight blind super-resolution model LESRCNN, the code has been open source
(03) Latest! Comprehensive comparative study of image denoising
(04) Without user input, Adobe proposes new method for automatic high-quality image synthesis
(05) Researchers at NVIDIA Research propose an adaptive discriminator enhancement mechanism that significantly stabilizes training in limited data environments.
(06) Photos become cartoon style in seconds! Teach you to use PaddleGAN to quickly generate your exclusive cartoon avatar
(07) and accurately generate Fake faces! Amazon's new GAN model gives you all-round beauty without dead ends
(08) The postdoctoral sister has upgraded the "two-dimensional wife generator"! AniGAN: This time you can specify the style of painting
(09) Training GANs 10 lessons I learned in a year
(10) Covering 18+ SOTA GAN implementations, this open source project PyTorch library fire
(11) 6ms EfficientDeRain: Inspiring simple and efficient rain removal algorithm
(12) PULSE: An image super-resolution algorithm based on implicit space
(13) Solve the problem that previous single-image super-resolution algorithms only work well on synthetic data and cannot be applied to real scenes. It can generalize to different cameras without training on images from a specific type of camera. Exploiting Raw Images for Real-Scene Super-Resolution Unit | Carnegie Mellon University, SenseTime, University of California Paper | https://arxiv.org/pdf/2102.01579.pdf Code | https://www.dropbox.com/s/a66iuwoswul65da/RawSR_PAMI20.zip?dl=0 Home | https://sites.google.com/ view/xiangyuxu/rawsr_pami (14) The next generation
locker room! A virtual fitting application made by a foreign designer. 2D joint point tracking based on OpenPose, rendered with Houdini special effects.
Source: https://80.lv/articles/next-gen-dressing-room-with-markerless-tracking-in-houdini/

(15) #GAN #WACV2021
SinGAN-GIF can generate samples of any aspect ratio, perform super-resolution, change the time frame rate, and can be used for video editing applications.
SinGAN-GIF: Learning a Generative Video Model From a Single GIF
Author | Rajat Arora, Yong Jae Lee
Unit | University of California, Davis Paper
|
https://openaccess.thecvf.com/content/WACV2021/papers/Arora_SinGAN-GIF_Learning_a_Generative_Video_Model_From_a_Single _GIF_WACV_2021_paper.pdf
home page | https://rajat95.github.io/singan-gif/

(16) WACV 2021 Paper Inventory - Image Quality

(17) Morph-UGATIT: An image translation method that supports progressive domain migration
(18) An extremely fast video frame interpolation method recently proposed by UC San Diego, CMU, and Facebook is 384 times faster than the previous most accurate method and 23 times faster than the previous fastest 8-fold interpolation method. This video is a slow motion image obtained by using this method. The code will be open source.
FLAVR: Flow-Agnostic Video Representations for Fast Frame Interpolation
units | University of California, San Diego; Carnegie Mellon University; Facebook AI
paper | https://arxiv.org/abs/2012.08512
github | coming
home | https://tarun005.github.io/FLAVR/

(19) The Neural Body algorithm invented by Zhejiang University and other scholars can output 3D human body and new angle views by inputting multi-angle videos. Paper | https://arxiv.org/pdf/2012.15838.pdf Code | https://github.com/zju3dv/neuralbody (coming soon)

(20) CVPR 2021 Accepted Paper: AdCo Contrastive Learning Based on Adversarial

(21) The first DNN solution to employ both sensor data and images for video stabilization. Deep Online Fused Video Stabilization Unit | University of Wisconsin-Madison, Google Paper | https://arxiv.org/abs/2102.01279 Code | Coming Home | https://zhmeishi.github.io/dvs/

(22) The Neural Body algorithm invented by Zhejiang University and other scholars can output 3D human body and new angle views by inputting multi-angle videos. Paper | https://arxiv.org/pdf/2012.15838.pdf Code | https://github.com/zju3dv/neuralbody (coming soon)

(23) TIP 2021 paper: Joint realization of multi-exposure image fusion and super-resolution
(24) CVPR 2021 Oral｜Silky 3D effects can be rendered in real time with only static images
(25) Intellectual interest丨Real-time style migration, running on mobile terminals, and face special effects have a new way
(26) CVPR 2021 | HKUST: How to use flash images to remove reflections?
(27) CVPR 2021 | Neighbor2Neighbor: A method for training arbitrary noise reduction networks with only noisy images
(28) Multifunctional image super-resolution model: Asymmetric convolutional neural network for blind image super-resolution
(29) A breakthrough in deep image restoration
(30) GANSpace: Discovering Interpretable GAN Controls
(31) TIP2021 | Multi-level feature fusion network in video super-resolution
(32) rt Flow: Unbiased Image Style Transfer via Reversible Neural Flows proposes ArtFlow to prevent content leaks in the general style transfer process. ArtFlow consists of a reversible neural flow and an unbiased feature transfer module. It supports both forward and backward reasoning, and operates using a projection-transfer-restore scheme. ArtFlow achieves comparable performance to state-of-the-art style transfer methods while avoiding content leaks.
(33)极市直播丨邓欣：TIP 2021-多曝光图像融合及超分辨的联合实现方法
（34）Weather GAN：实现晴、阴、雾、雨、雪之间的天气状况自由迁移
（35）CVPR 2021 | 五官画风都能改，用无监督方法控制 GAN (附源码) -周博磊团队
（36）CVPR 2021｜Neighbor2Neighbor：无需干净图像的自监督图像降噪
（37）图像反光能被一键去除了？港科大开源RFC，仅用一个操作，强反光也能完美去除｜CVPR2021
（38）你好，这是微视AI还原的李焕英
（39）有限数据来训练GAN的一种思路
（40）揭秘腾讯微视人脸技术「黑科技」，基于GAN的人脸魔法特效
（41）添加链接描述
（42）CVPR2021｜超分性能不降低，计算量降低50%，董超等人提出加速图像超分的ClassSR
（43）RealSR性能大幅提升！旷视+快手+电子科大联合提出“先发散再收敛”的D2CSR
（44）仅需2张图！AI便可生成完整运动过程
（45）PornHub 用独家数据集！修复了百年前的电影…
（46）新垣结衣夫妇的孩子会长啥样？我用BabyGAN预测试试…
（47）just! AMD releases new super-resolution technology FSR: N cards can also be used
(48) Transformer makes another breakthrough! ETH proposes: Video super-resolution Transformer
(49) Cai Xukun x special xxx dream linkage! This artifact allows images to imitate human movements in real time
(50) Everyone can use the second dimension! This GAN network allows Miss Sister to generate different styles of anime images! Changeable skin color and hairstyle
(51) This AI artifact brings father back to 18 years old! GAN model generates anime portraits in 130 milliseconds! (60)

Deepfake text version turned out: AI high imitation of your handwriting only needs 1 word!
(61) [Open source] Discussion on font generation based on image background, human body pose prediction, key point detection, super-resolution, etc. (62
) https://intel-isl.github.io/PhotorealismEnhancement/
(63) Image filling is not afraid of large areas! MSRA et al. proposed co-modulation generative confrontation network
(54) CVPR 2021 Oral | GLEAN: High-magnification image super-resolution based on implicit generation library
(55) beats the pack! 2021 NTIRE @CVPR 2021 Triple Crown and One Sub-Video Super Score Solution: BasicVSR++