Normalized and Geometry-Aware Self-Attention Network for Image Captioning
重点在自注意力机制的image captioning方法上。 现有的Self-Attention方法作者认为存在两个问题： 一个是：Internal Covariate Shift 我的理解就是输入分布不一样 解决办法就是Normalization。 原来的Transformer当中也是有Normalization的，但是作者认为原来的做法不够好： 翻译过来，就是要把norm放到自注意力模块里面
相关文章
- 1. Reflective Decoding Network for Image Captioning论文阅读
- 2. Self-critical Sequence Training for Image Captioning
- 3. 《Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering》
- 4. Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering
- 5. (Paper Reading)Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering
- 6. CVPR 2018 Oral：Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering
- 7. Pose-Normalized Image Generation for Person Re-identification (note)
- 8. 论文笔记：Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering
- 9. Ncut算法（Normalized cuts and image segmentation）
- 10. PEPSI++: Fast and Lightweight Network for Image Inpainting | 简记
- 更多相关文章...
- • RSS
元素 - RSS 教程 - • ASP.NET Image 控件 - ASP.NET 教程
- • RxJava操作符（七）Conditional and Boolean
- • Flink 数据传输及反压详解
相关标签/搜索
每日一句
-
每一个你不满意的现在，都有一个你没有努力的曾经。
欢迎关注本站公众号,获取更多信息
相关文章
- 1. Reflective Decoding Network for Image Captioning论文阅读
- 2. Self-critical Sequence Training for Image Captioning
- 3. 《Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering》
- 4. Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering
- 5. (Paper Reading)Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering
- 6. CVPR 2018 Oral：Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering
- 7. Pose-Normalized Image Generation for Person Re-identification (note)
- 8. 论文笔记：Bottom-Up and Top-Down Attention for Image Captioning and Visual Question Answering
- 9. Ncut算法（Normalized cuts and image segmentation）
- 10. PEPSI++: Fast and Lightweight Network for Image Inpainting | 简记