Haonan Zhang
Haonan Zhang
Yes, the predicted object labels are not used for captioning task.
Hello, I got the same issues, can you send me a copy of TGIF-QA (visual and textual features)? Thank you very much! This is my email ([[email protected]]([email protected])).
Thanks for your attention to our work. This [link ](https://github.com/albanie/collaborative-experts/blob/master/misc/datasets/msvd/README.md)may provide some help. 
 我点击选项卡之后会出错，不能选择，请问是为啥呀
 还有一个问题就是在群聊里面无法进入图片创作模式，会报错
yes, doesn't work again
got it. thanks~
Solved by removing this line https://github.com/RayeRen/acad-homepage.github.io/blob/2cc1577eeaf2f74dede6d016a70722dbd409ea2f/_pages/about.md?plain=1#L4
您好，可以参考这篇论文[1]，他的repo里面提供了特征抽取方案(hdf5 file): [Link](https://github.com/facebookresearch/grid-feats-vqa) [1] Jiang, H., Misra, I., Rohrbach, M., Learned-Miller, E., & Chen, X. (2020). In defense of grid features for visual question answering. In Proceedings of the IEEE/CVF...
您好，由于本项目是离线视觉特征训练和测试，如果要实际部署，需要修改成端到端推理。有几点建议包括但不限于：1. 修改dataloader，从而对输入图像进行online的特征抽取，见#12 ，2. 下面的代码包含了模型输出解码和评测，```gts```如果不需要你可以删除掉，只要模型当前的输出```gen_i```即可。 https://github.com/zchoi/S2-Transformer/blob/c2f56c5715fc44e4677f546e5bb7c36d57909374/test_transformer.py#L33-L37