search

qizekun / ReCon

[ICML 2023] Contrast with Reconstruct: Contrastive 3D Representation Learning Guided by Generative Pretraining
https://arxiv.org/abs/2302.02318
MIT License
120 stars 13 forks source link

Why is the Leaderboard result about zero-shot classification lower than paper #11

Closed TangYuan96 closed 10 months ago

TangYuan96 commented 10 months ago

First, thanks for sharing so good work!

Why is the Leaderboard result about zero-shot classification lower than paper?

zero-shot classification on MN10： 81.6（ in paper）> 75.6 (inLeaderboard )

zero-shot classification on MN40： 66.8（ in paper）> 61.7 (inLeaderboard )

Look forward to your reply~

qizekun commented 10 months ago

I'm very sorry for the inconvenience caused. MN40: 66.8% and MN10: 81.6% were zero-shot tested on the complete ModelNet dataset (i.e., train+test), but the more appropriate comparison should actually be done only on the test set. We have made the necessary corrections in the latest arXiv paper version and on the Leaderboard.

Also, we noticed your previous issue and have uploaded ModelNet10 to Google Drive for you to use.

We sincerely appreciate your interest in our work!