【CVPR2018】论文整理
转载自：https://blog.csdn.net/leviopku/article/details/80783301
CVPR作为CV界最受关注的三大顶会之一，每一个CVer都应该好好关注CVPR的论文。CVPR2018在今年6月18日-22日在美国盐湖城举行。
先介绍一下CVPR2018的一些数据：
- 今年一共收到3309篇文章，其中979篇被录用。投录比约为29.5%。
- 收录论文按专家评分，分为三个层次：Poster, Spotlight, Oral。
- Spotlight(亮点论文)一共有224篇，占收录论文(224/979)的22.88%。
- Oral(演示论文)一共有70篇，占收录论文(70/979)的7.1%。
用一张韦恩图表示收录文章占比：
所以说，不光中篇CVPR难，中篇spotlight更难，中篇oral基本可以说是灰常难了。就这么说吧，今年国内所有高校加起来中的CVPR oral是个位数。
当然，最牛的还是Best paper 和best student paper，只会分别选出1篇。
今年的best paper给了来自Stanford和Berkeley的合作论文，论文标题为：
| Taskonomy: Disentangling Task Transfer Learning |
|---|
| 下载地址为：https://arxiv.org/abs/1804.08328 |
最佳学生论文来自CMU，标题为：
| Total Capture: A 3D Deformation Model for Tracking Faces, Hands, and Bodies |
|---|
| 下载地址为：https://arxiv.org/abs/1801.01615v1 |
当然，就像奥斯卡颁奖一样，最佳论文奖提名也可以突出文章质量很高。今年四篇最佳论文提名奖如下：
| 标题 | 第一单位 | 下载地址 |
|---|---|---|
| Deep_Learning_of_Graph_Matching | Lund University | http://openaccess.thecvf.com /content_cvpr_2018 /CameraReady/1830.pdf |
| SPLATNet: Sparse Lattice Networks for Point Cloud Processing | UMass Amherst | https://arxiv.org/pdf/1802.08275.pdf |
| CodeSLAM-learning a Compact, Optimisable Representation for Dense Visual SLAM | 帝国理工 | https://arxiv.org/pdf/1804.00874.pdf |
| Efficient Optimization for Rank-based Loss Functions | IIIT Hyderabad | https://arxiv.org/pdf/1604.08269.pdf |
所以，客观认为的论文含金量是：
best paper (2篇) > honorable mention(提名奖 4篇) > Oral (70篇) > Spotlight(224篇) > poster(其他)
CVPR2018虽好，可不要贪杯，一共有979篇，每天看1篇也得看3年，待你看完之日也是算法过时之时。所以，给各位CVer(包括自己)一些建议：
- 从高质量论文开始看，至少优先看spotlight或者oral论文。
- 在自己的领域找论文看，别想做什么CVPR的集大成者，如果你是CVPR oral大神，那么当我这条没说过。
- 哪里有CVPR论文分享会就去听，听原作者自己讲一个小时，比自己看一礼拜更管用。如果没有现场版，看看视频也是好的。
最后
附上68篇oral论文标题：（文末有下载链接）
| 1 | DensePose: Multi-Person Dense Human Pose Estimation In The Wild |
|---|---|
| 2 | Context Encoding for Semantic Segmentation |
| 3 | Augmented Skeleton Space Transfer for Depth-based Hand Pose Estimation |
| 4 | Semi-parametric Image Synthesis |
| 5 | Practical Block-wise Neural Network Architecture Generation |
| 6 | Are You Talking to Me? Reasoned Visual Dialog Generation through Adversarial Learning |
| 7 | PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume |
| 8 | Illuminant Spectra-based Source Separation Using Flash Photography |
| 9 | SPLATNet: Sparse Lattice Networks for Point Cloud Processing |
| 10 | Total Capture: A 3D Deformation Model for Tracking Faces, Hands, and Bodies |
| 11 | Deep Layer Aggregation |
| 12 | Left-Right Comparative Recurrent Model for Stereo Matching |
| 13 | Analytic Expressions for Probabilistic Moments of PL-DNN with Gaussian Input |
| 14 | An Analysis of Scale Invariance in Object Detection - SNIP |
| 15 | Finding Tiny Faces in the Wild with Generative Adversarial Network |
| 16 | Taskonomy: Disentangling Task Transfer Learning |
| 17 | High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs |
| 18 | Finding “It”: Weakly-Supervised Reference-Aware Visual Grounding in Instructional Video |
| 19 | Unsupervised Discovery of Object Landmarks as Structural Representations |
| 20 | Rotation Averaging and Strong Duality |
| 21 | Im2Flow: Motion Hallucination from Static Images for Action Recognition |
| 22 | Group Consistent Similarity Learning via Deep CRFs for Person Re-Identification |
| 23 | 3D-RCNN: Instance-level 3D Scene Understanding via Render-and-Compare |
| 24 | Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering |
| 25 | Context Contrasted Feature and Gated Multi-scale Aggregation for Scene Segmentation |
| 26 | Squeeze-and-Excitation Networks |
| 27 | DoubleFusion: Real-time Capture of Human Performance with Inner Body Shape from a Single Depth Sensor |
| 28 | Learning to Find Good Correspondences |
| 29 | Actor and Action Video Segmentation from a Sentence |
| 30 | Maximum Classifier Discrepancy for Unsupervised Domain Adaptation |
| 31 | Detail-Preserving Pooling in Deep Networks |
| 32 | Convolutional Neural Networks with Alternately Updated Clique |
| 33 | Deep Learning of Graph Matching |
| 34 | Synthesizing Images of Humans in Unseen Poses |
| 35 | Neural Inverse Kinematics for Unsupervised Motion Retargetting |
| 36 | Direction-aware Spatial Context Features for Shadow Detection |
| 37 | Density Adaptive Point Set Registration |
| 38 | Hybrid Camera Pose Estimation |
| 39 | Relation Networks for Object Detection |
| 40 | Revisiting Salient Object Detection: Simultaneous Detection, Ranking, and Subitizing of Multiple Salient Objects |
| 41 | Im2Pano3D: Extrapolating 360 Structure and Semantics Beyond the Field of View |
| 42 | Polarimetric Dense Monocular SLAM |
| 43 | Wasserstein Introspective Neural Networks |
| 44 | The Perception-Distortion Tradeoff |
| 45 | Discriminative Learning of Latent Features for Zero-Shot Recognition |
| 46 | Photometric Stereo in Participating Media Considering Shape-Dependent Forward Scatter |
| 47 | Fast and Furious: Real Time End-to-End 3D Detection, Tracking and Motion Forecasting with a Single Convolutional Net |
| 48 | Trapping Light for Time of Flight |
| 49 | Feature Space Transfer for Data Augmentation |
| 50 | Self-supervised Multi-level Face Model Learning for Monocular Reconstruction at over 250Hz |
| 51 | CodeSLAM — Learning a Compact, Optimisable Representation for Dense Visual SLAM |
| 52 | FlipDial: A Generative Model for Two-Way Visual Dialogue |
| 53 | OATM: Occlusion Aware Template Matching by Consensus Set Maximization |
| 54 | Surface Networks |
| 55 | VirtualHome: Simulating Household Activities via Programs |
| 56 | Egocentric Activity Recognition on a Budget |
| 57 | Improved Fusion of Visual and Language Representations by Dense Symmetric Co-Attention for Visual Question Answering |
| 58 | Efficient Optimization for Rank-based Loss Functions |
| 59 | MakeupGAN: Makeup Transfer via Cycle-Consistent Adversarial Networks |
| 60 | Revisiting Deep Intrinsic Image Decompositions |
| 61 | StarGAN: Unified Generative Adversarial Networks for Controllable Multi-Domain Image-to-Image Translation |
| 62 | Ordinal Depth Supervision for 3D Human Pose Estimation |
| 63 | Multi-Cell Classification by Convolutional Dictionary Learning with Class Proportion Priors |
| 64 | Accurate and Diverse Sampling of Sequences based on a “Best of Many” Sample Objective |
| 65 | MapNet: An Allocentric Spatial Memory for Mapping Environments |
| 66 | A Globally Optimal Solution to the Non-Minimal Relative Pose Problem |
| 67 | A Volumetric Descriptive Network for 3D Object Synthesis |
| 68 | Learning Face Age Progression: A Pyramid Architecture of GANs |