动态transformer

动态Transformer是一种针对每个样本选择合适数目的token进行表征的视觉模型框架。该框架的实现可以参考GitHub上的开源代码。在动态Transformer中，通过在前向推理过程中根据输入图像尺寸动态生成mask，实现了对Swin-Transformer的动态输入尺寸的支持。这种动态生成mask的计算量较低，并且不涉及插值等操作。通过灵活地选择token数量，动态Transformer能够更好地适应不同样本的特征表达需求。<span class="em">1</span><span class="em">2</span><span class="em">3</span>
#### 引用[.reference_title]
- *1* *3* [不是所有图像都值 16x16 个词，可变序列长度的动态 Transformer 来了！](https://blog.csdn.net/qq_33431368/article/details/117857453)[target="_blank" data-report-click={"spm":"1018.2226.3001.9630","extra":{"utm_source":"vip_chatgpt_common_search_pc_result","utm_medium":"distribute.pc_search_result.none-task-cask-2~all~insert_cask~default-1-null.142^v93^chatsearchT3_2"}}] [.reference_item style="max-width: 50%"]
- *2* [以动制动 | Transformer 如何处理动态输入尺寸](https://blog.csdn.net/qq_39967751/article/details/123666686)[target="_blank" data-report-click={"spm":"1018.2226.3001.9630","extra":{"utm_source":"vip_chatgpt_common_search_pc_result","utm_medium":"distribute.pc_search_result.none-task-cask-2~all~insert_cask~default-1-null.142^v93^chatsearchT3_2"}}] [.reference_item style="max-width: 50%"]
[ .reference_list ]