Qingzhong Wang

Solved. Just install zhusuan 0.3.1 and correct 2 errors based on the error information. If someone use python2, you should change the protocol of the pickle file in the obj_vectors...

> Hi @ruotianluo. Thanks for your codebase. > I trained AoA model using your code and your params config. however the performance i get is worser than that paper report....

> @qingzwang 我找不到bug5555555555555555555555555。用作者的code吧。。。。。似乎是我改了[30a0e7b](https://github.com/ruotianluo/self-critical.pytorch/commit/30a0e7b4b572b1a48d64d2f7f3574493fd3c7d56，但是我始终没找到到底是哪改错了。) 作者code也是基于你早期self-critical，没发现太多不同，好难啊

> 你要不帮我看看？我自己看可能会盲目。我基本上确定是由于这个commit导致的你也可以试试，回滚到这个commit之前就对了 我先对比一下这俩版本

@ruotianluo @upccpu @zhangxuying1004 I have tested the publicly available [code](https://github.com/husthuaan/AoANet) released by the author and this repository. Both of them could obtain CIDEr of 118.0, but cannot achieve 119.8 reported...

> 具体有哪些中文captioning数据集呢？或者有公开文档可以看看么？感觉Taisu之类的数据集噪声很大，直接用这个数据集做captioning看不到效果。