Unsupervised-Learning-of-Robust-Spectral-Shape-Matching
Unsupervised-Learning-of-Robust-Spectral-Shape-Matching copied to clipboard
dongliangcao
Result
I found that I can't reproduce the result in the paper, and the released pretrained model is also work badly. So , could you give some more instruction about this?
I went through your github code. 1.I only changed the non-isometric in the option.py file of the smal dataset to false (because I found that it works a little better this way) 2. and then I downloaded the dataset you provided and processed it through the dataset processing code you given 3. and then trained it according to your train.py file (no test enhancement) However, the SMAL result was 6.3, a slight decrease from 5.8 in the paper. What gpu did you use for your experiments? Because I found that different GPUs may slightly affect the result.
------------------ 原始邮件 ------------------ 发件人: "dongliangcao/Unsupervised-Learning-of-Robust-Spectral-Shape-Matching" @.>; 发送时间: 2024年6月5日(星期三) 下午5:59 @.>; @.@.>; 主题: Re: [dongliangcao/Unsupervised-Learning-of-Robust-Spectral-Shape-Matching] Result (Issue #8)
Could you specify the problem you have and list the steps you tried?
— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
Hello, I would like to ask a question. When you trained using the FAUST dataset and then tested the FAUST, SCAPE, and SHPEC'19 datasets separately, did you directly use the parameters corresponding to the FAUST, SCAPE, and SHPEC'19 datasets in the "test" directory of the option file, only modifying the weights used for FAUST? Do you need to modify other contents? The data I obtained does not reach the effect described in the paper.
