Shizhe Chen

icon indicating a website link https://cshizhe.github.io/ icon indicating an email [email protected]

icon company INRIA icon location Paris

Results 10 comments of Shizhe Chen

Fails when training from scratch

Thank you. I tried the first approach. The best performance I got on scannetv2 is mAP/mAP50/mAP25 = 0.435/0.642/0.772. Could you reproduce the results when training from scratch on scannetv2? BTW,...

Error(s) in loading state_dict for CMAPolicy:

> Do you have an idea why the performance is much lower than the result listed [here](https://github.com/jacobkrantz/VLN-CE/blob/d2f73730a4b9df551388ff78d378956043b1635f/vlnce_baselines/config/r2r_baselines/README.md)?

Could you please share the running scripts for IL+RL training from scratch?

Hi, you could simply modify the configurations to train HAMT from scratch, e.g. no fix embeddings and no init from existing checkpoints.

I found that only the positive examples are given attention in the paper, is there any data leakage?

Could you explain more about that? The proposed attention will be applied for each candidate pair.

Val-unseen SR&SPL in pretrain and finetune

How do you select the best checkpoint? It seems to be a similar issue as [this](https://github.com/cshizhe/VLN-DUET/issues/13).

Is there a script to generate image captions like Figure 5 in the paper

Thanks for your interests. I am preparing a demo and will release it soon.

where is the pretrained models and codes to extract features for new images?

> Thanks for your great work ! > I was interested in your work! > > ![sssdddd](https://user-images.githubusercontent.com/55009815/87310873-227f1c80-c551-11ea-869c-f87d3f6e743b.png) > > > We also provide pretrained models and codes to extract features...

where is the pretrained models and codes to extract features for new images?

> same question. Is there an avaliable pretrained asg2cap model provided? Thanks! I am preparing a demo and will release it soon.

R2R Test Unseen

You should submit the results to the R2R leaderboard: https://eval.ai/web/challenges/challenge-page/606/leaderboard/1683

Would you mind sharing PCD files? Thanks

Thank you for sharing the point cloud data. I would like to download it, but it seems that you should add my email address to the box.net in order to...