CVPR2019年熱門論文及開原始碼分享

    CVPR是國際上首屈一指的年度計算機視覺會議，由主要會議和幾個共同舉辦的研討會和短期課程組成。憑藉其高品質和低成本，為學生，學者和行業研究人員提供了難得的交流學習的機會。

    CVPR2019將於6月16日至6月20日，在美國加州長灘舉行。會議首頁：http://cvpr2019.thecvf.com/

    本文整理了本屆會議最熱門的100篇論文，以及論文開源的Github程式碼地址（按照Github star數目排序）。

論文串列

    Detect-To-Retrieve: Efficient Regional Aggregation for Image Search

    Github地址：https://github.com/tensorflow/models/tree/master/research/delf

    A Style-Based Generator Architecture for Generative Adversarial Networks

    Github地址：https://github.com/NVlabs/stylegan

    Semantic Image Synthesis With Spatially-Adaptive Normalization

    Github地址：https://github.com/NVlabs/SPADE

    ArcFace: Additive Angular Margin Loss for Deep Face Recognition

    Github地址：https://github.com/deepinsight/insightface

    Deformable ConvNets V2: More Deformable, Better Results

    Github地址：https://github.com/msracver/Deformable-ConvNets

    Bag of Tricks for Image Classification with Convolutional Neural Networks

    Github地址：https://github.com/dmlc/gluon-cv

    Towards VQA Models That Can Read

    Github地址：https://github.com/facebookresearch/pythia

    DSFD: Dual Shot Face Detector

    Github地址：https://github.com/TencentYoutuResearch/FaceDetection-DSFD

    Fast Online Object Tracking and Segmentation: A Unifying Approach

    Github地址：https://github.com/foolwood/SiamMask

    Deep High-Resolution Representation Learning for Human Pose Estimation

    Github地址：https://github.com/leoxiaobin/deep-high-resolution-net.pytorch

    Mask Scoring R-CNN

    Github地址：https://github.com/zjhuang22/maskscoring_rcnn

    Joint Discriminative and Generative Learning for Person Re-Identification

    Github地址：https://github.com/layumi/Person_reID_baseline_pytorch

    Self-Supervised GANs via Auxiliary Rotation Loss

    Github地址：https://github.com/google/compare_gan

    PartNet: A Large-Scale Benchmark for Fine-Grained and Hierarchical Part-Level 3D Object Understanding

    Github地址：https://github.com/yangyanli/PointCNN

    ABC: A Big CAD Model Dataset for Geometric Deep Learning

    Github地址：https://github.com/yangyanli/PointCNN

    Dual Attention Network for Scene Segmentation

    Github地址：https://github.com/junfu1115/DANet

    DeepFashion2: A Versatile Benchmark for Detection, Pose Estimation, Segmentation and Re-Identification of Clothing Images

    Github地址：https://github.com/switchablenorms/DeepFashion2

    Bottom-Up Object Detection by Grouping Extreme and Center Points

    Github地址：https://github.com/xingyizhou/ExtremeNet

    3D Human Pose Estimation in Video With Temporal Convolutions and Semi-Supervised Training

    Github地址：https://github.com/facebookresearch/VideoPose3D

    HAQ: Hardware-Aware Automated Quantization With Mixed Precision

    Github地址：https://github.com/MIT-HAN-LAB/ProxylessNAS

    Fast Interactive Object Annotation With Curve-GCN

    Github地址：https://github.com/fidler-lab/curve-gcn

    ATOM: Accurate Tracking by Overlap Maximization

    Github地址：https://github.com/martin-danelljan/ECO

    Shape Robust Text Detection With Progressive Scale Expansion Network

    Github地址：https://github.com/whai362/PSENet

    Data-Driven Neuron Allocation for Scale Aggregation Networks

    Github地址：https://github.com/lxtGH/OctaveConv_pytorch

    ESPNetv2: A Light-Weight, Power Efficient, and General Purpose Convolutional Neural Network

    Github地址：https://github.com/osmr/imgclsmob

    MnasNet: Platform-Aware Neural Architecture Search for Mobile

    Github地址：https://github.com/osmr/imgclsmob

    Selective Kernel Networks

    Github地址：https://github.com/osmr/imgclsmob

    Deep Plug-And-Play Super-Resolution for Arbitrary Blur Kernels

    Github地址：https://github.com/cszn/DPSR

    Recurrent MVSNet for High-Resolution Multi-View Stereo Depth Inference

    Github地址：https://github.com/YoYo000/MVSNet

    PointRCNN: 3D Object Proposal Generation and Detection From Point Cloud

    Github地址：https://github.com/sshaoshuai/PointRCNN

    DenseFusion: 6D Object Pose Estimation by Iterative Dense Fusion

    Github地址：https://github.com/j96w/DenseFusion

    Feature Denoising for Improving Adversarial Robustness

    Github地址：https://github.com/facebookresearch/ImageNet-Adversarial-Training

    Bounding Box Regression With Uncertainty for Accurate Object Detection

    Github地址：https://github.com/yihui-he/softer-NMS

    High-Level Semantic Feature Detection: A New Perspective for Pedestrian Detection

    Github地址：https://github.com/liuwei16/CSP

    A Dataset and Benchmark for Large-Scale Multi-Modal Face Anti-Spoofing

    Github地址：https://github.com/SoftwareGift/FeatherNets_Face-Anti-spoofing-Attack-Detection-Challenge-CVPR2019

    Stereo R-CNN Based 3D Object Detection for Autonomous Driving

    Github地址：https://github.com/HKUST-Aerial-Robotics/Stereo-RCNN

    From Recognition to Cognition: Visual Commonsense Reasoning

    Github地址：https://github.com/rowanz/r2c

    Deeper and Wider Siamese Networks for Real-Time Visual Tracking

    Github地址：https://github.com/researchmm/SiamDW

    Auto-DeepLab: Hierarchical Neural Architecture Search for Semantic Image Segmentation

    Github地址：https://github.com/MenghaoGuo/AutoDeeplab

    PlaneRCNN: 3D Plane Detection and Reconstruction From a Single Image

    Github地址：https://github.com/art-programmer/PlaneNet

    Precise Detection in Densely Packed Scenes

    Github地址：https://github.com/eg4000/SKU110K_CVPR19

    Self-Supervised Learning of 3D Human Pose Using Multi-View Geometry

    Github地址：https://github.com/mkocabas/EpipolarPose

    PVNet: Pixel-Wise Voting Network for 6DoF Pose Estimation

    Github地址：https://github.com/zju3dv/pvnet

    4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks

    Github地址：https://github.com/StanfordVL/MinkowskiEngine

    Progressive Pose Attention Transfer for Person Image Generation

    Github地址：https://github.com/tengteng95/Pose-Transfer

    PifPaf: Composite Fields for Human Pose Estimation

    Github地址：https://github.com/vita-epfl/openpifpaf

    PointPillars: Fast Encoders for Object Detection From Point Clouds

    Github地址：https://github.com/nutonomy/second.pytorch

    Pseudo-LiDAR From Visual Depth Estimation: Bridging the Gap in 3D Object Detection for Autonomous Driving

    Github地址：https://github.com/mileyan/pseudo_lidar

    UPSNet: A Unified Panoptic Segmentation Network

    Github地址：https://github.com/uber-research/UPSNet

    Mode Seeking Generative Adversarial Networks for Diverse Image Synthesis

    Github地址：https://github.com/HelenMao/MSGAN

    Image Super-Resolution by Neural Texture Transfer

    Github地址：https://github.com/ZZUTK/SRNTT

    PointConv: Deep Convolutional Networks on 3D Point Clouds

    Github地址：https://github.com/DylanWusee/pointconv

    Decoders Matter for Semantic Segmentation: Data-Dependent Decoding Enables Flexible Feature Aggregation

    Github地址：https://github.com/LinZhuoChen/DUpsampling

    Occupancy Networks: Learning 3D Reconstruction in Function Space

    Github地址：https://github.com/LMescheder/Occupancy-Networks

    DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation

    Github地址：https://github.com/LMescheder/Occupancy-Networks

    Learning Implicit Fields for Generative Shape Modeling

    Github地址：https://github.com/LMescheder/Occupancy-Networks

    Toward Convolutional Blind Denoising of Real Photographs

    Github地址：https://github.com/GuoShi28/CBDNet

    Competitive Collaboration: Joint Unsupervised Learning of Depth, Camera Motion, Optical Flow and Motion Segmentation

    Github地址：https://github.com/anuragranj/cc

    Recurrent Back-Projection Network for Video Super-Resolution

    Github地址：https://github.com/alterzero/RBPN-PyTorch

    Pluralistic Image Completion

    Github地址：https://github.com/lyndonzheng/Pluralistic-Inpainting

    Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-Identification

    Github地址：https://github.com/zhunzhong07/ECN

    ScratchDet: Training Single-Shot Object Detectors From Scratch

    Github地址：https://github.com/KimSoybean/ScratchDet

    Rethinking Knowledge Graph Propagation for Zero-Shot Learning

    Github地址：https://github.com/cyvius96/adgpm

    Feedback Network for Image Super-Resolution

    Github地址：https://github.com/Paper99/SRFBN_CVPR19

    Associatively Segmenting Instances and Semantics in Point Clouds

    Github地址：https://github.com/WXinlong/ASIS

    Unsupervised Person Re-Identification by Soft Multilabel Learning

    Github地址：https://github.com/KovenYu/MAR

    Detailed Human Shape Estimation From a Single Image by Hierarchical Mesh Deformation

    Github地址：https://github.com/zhuhao-nju/hmd

    AOGNets: Compositional Grammatical Architectures for Deep Learning

    Github地址：https://github.com/xilaili/AOGNet

    GANFIT: Generative Adversarial Network Fitting for High Fidelity 3D Face Reconstruction

    Github地址：https://github.com/barisgecer/ganfit

    Data Augmentation Using Learned Transformations for One-Shot Medical Image Segmentation

    Github地址：https://github.com/xamyzhao/brainstorm

    Meta-Learning With Differentiable Convex Optimization

    Github地址：https://github.com/kjunelee/MetaOptNet

    Single-Image Piece-Wise Planar 3D Reconstruction via Associative Embedding

    Github地址：https://github.com/svip-lab/PlanarReconstruction

    Style Transfer by Relaxed Optimal Transport and Self-Similarity

    Github地址：https://github.com/nkolkin13/STROTSS

    Class-Balanced Loss Based on Effective Number of Samples

    Github地址：https://github.com/richardaecn/class-balanced-loss

    Panoptic Segmentation

    Github地址：https://github.com/cocodataset/panopticapi

    Revisiting Self-Supervised Visual Representation Learning

    Github地址：https://github.com/google/revisiting-self-supervised

    Real-Time Self-Adaptive Deep Stereo

    Github地址：https://github.com/CVLAB-Unibo/Real-time-self-adaptive-deep-stereo

    D2-Net: A Trainable CNN for Joint Description and Detection of Local Features

    Github地址：https://github.com/mihaidusmanu/d2-net

    Hardness-Aware Deep Metric Learning

    Github地址：https://github.com/wzzheng/HDML

    Animating Arbitrary Objects via Deep Motion Transfer

    Github地址：https://github.com/AliaksandrSiarohin/monkey-net

    Learning Attraction Field Representation for Robust Line Segment Detection

    Github地址：https://github.com/cherubicXN/afm_cvpr2019

    Fast Human Pose Estimation

    Github地址：https://github.com/yuanyuanli85/Fast_Human_Pose_Estimation_Pytorch

    CrowdPose: Efficient Crowded Scenes Pose Estimation and a New Benchmark

    Github地址：https://github.com/Jeff-sjtu/CrowdPose

    FSA-Net: Learning Fine-Grained Structure Aggregation for Head Pose Estimation From a Single Image

    Github地址：https://github.com/shamangary/FSA-Net

    Photometric Mesh Optimization for Video-Aligned 3D Object Reconstruction

    Github地址：https://github.com/chenhsuanlin/photometric-mesh-optim

    RVOS: End-To-End Recurrent Network for Video Object Segmentation

    Github地址：https://github.com/imatge-upc/rvos

    Pose2Seg: Detection Free Human Instance Segmentation

    Github地址：https://github.com/liruilong940607/Pose2Seg

    MUREL: Multimodal Relational Reasoning for Visual Question Answering

    Github地址：https://github.com/Cadene/murel.bootstrap.pytorch

    3D-SIS: 3D Semantic Instance Segmentation of RGB-D Scans

    Github地址：https://github.com/Sekunde/3D-SIS

    PoseFix: Model-Agnostic General Human Pose Refinement Network

    Github地址：https://github.com/mks0601/PoseFix_RELEASE

    In Defense of Pre-Trained ImageNet Architectures for Real-Time Semantic Segmentation of Road-Driving Images

    Github地址：https://github.com/orsic/swiftnet

    FBNet: Hardware-Aware Efficient ConvNet Design via Differentiable Neural Architecture Search

    Github地址：https://github.com/JunrQ/NAS

    PointNetLK: Robust & Efficient Point Cloud Registration Using PointNet

    Github地址：https://github.com/hmgoforth/PointNetLK

    Relation-Shape Convolutional Neural Network for Point Cloud Analysis

    Github地址：https://github.com/Yochengliu/Relation-Shape-CNN

    Learning Linear Transformations for Fast Image and Video Style Transfer

    Github地址：https://github.com/sunshineatnoon/LinearStyleTransfer

    Strong-Weak Distribution Alignment for Adaptive Object Detection

    Github地址：https://github.com/VisionLearningGroup/DA_Detection

    Learning to Regress 3D Face Shape and Expression From an Image Without 3D Supervision

    Github地址：https://github.com/soubhiksanyal/RingNet

    Depth-Aware Video Frame Interpolation

    Github地址：https://github.com/baowenbo/DAIN

    Show, Control and Tell: A Framework for Generating Controllable and Grounded Captions

    Github地址：https://github.com/aimagelab/show-control-and-tell

    Tightness-Aware Evaluation Protocol for Scene Text Detection

    Github地址：https://github.com/Yuliang-Liu/TIoU-metric