【导读】本届CVPR共收到了6656篇有效投稿,接收1470篇,其接受率在逐年下降,今年接受率仅为22%。本文为大家整理了CVPR 2020 图神经网络(GNN)相关的比较有意思的值得阅读的10篇论文,供大家参考—分别是:点云分析、视频描述生成、轨迹预测、场景图生成、视频理解、3D语义分割、指代表达式推理、图像描述生成、图像处理、时空图等方向文章。
1. Grid-GCN for Fast and Scalable Point Cloud Learning
2. Object Relational Graph with Teacher-Recommended Learning for Video Captioning
3. Social-STGCNN: A Social Spatio-Temporal Graph Convolutional Neural Network for Human Trajectory Prediction
4. Unbiased Scene Graph Generation from Biased Training
5. Where Does It Exist: Spatio-Temporal Video Grounding for Multi-Form Sentences
6. Learning 3D Semantic Scene Graphs from 3D Indoor Reconstructions
7. Graph-Structured Referring Expression Reasoning in The Wild
8. Say As Y ou Wish: Fine-grained Control of Image Caption Generation with Abstract Scene Graphs
9. Semantic Image Manipulation Using Scene Graphs
10. Spatio-Temporal Graph for Video Captioning with Knowledge Distillation