目录
- [1] Bi-directional Relationship Inferring Network for Referring Image Segmentation
- [2] A Real-Time Cross-modality Correlation Filtering Method for Referring Expression Comprehension
- [3] Vision-Dialog Navigation by Exploring Cross-modal Memory
- [4] VQA with No Questions-Answers Training
- [5] Referring Image Segmentation via Cross-Modal Progressive Comprehension
- [6] Local-Global Video-Text Interactions for Temporal Grounding
- [7] Hypergraph Attention Networks for Multimodal Learning
- 总结
[1] Bi-directional Relationship Inferring Network for Referring Image Segmentation
- 卢湖川老师
- 已有方法：语言->视觉，没有视觉->语言。（->：指导）
[2] A Real-Time Cross-modality Correlation Filtering Method for Referring Expression Comprehension
- 北航刘偲、中山李冠斌
- 现有方法：两阶段（生成proposals、选最优proposal）比较慢
- 将相关滤波引入跨模态领域，用language feature当做kernel，在image feature上做相关滤波，得到响应图（bbox的中心），再回归w和h。
- 像极了SiamRPN，只不过一个branch改成了另一个模态。
[3] Vision-Dialog Navigation by Exploring Cross-modal Memory
- 跨模态记忆问题？
- 导航：只基于对话历史->加入视觉模块
[4] VQA with No Questions-Answers Training
- 不用answer就可以训练。
- 通过问题图，生成问题，生成的问题的答案没有意义。
[5] Referring Image Segmentation via Cross-Modal Progressive Comprehension
- 额，没太听懂。
[6] Local-Global Video-Text Interactions for Temporal Grounding
[7] Hypergraph Attention Networks for Multimodal Learning
总结
这次结束的超级快，一小时20分钟。