search

heartStrive / HandGCAT

9 stars 0 forks source link

About results #1

Open llllljw opened 7 months ago

llllljw commented 7 months ago

Hey, sorry to bother you, I'm following your work but I have some questions. I train the model on HO3D_V2 dataset and gain the pred70.zip, then I submit it to the competition, but the results are very poor, but I have not change codes.

heartStrive commented 7 months ago

Thank you for your interest. 2D keypoint detection can significantly affect model performance, so I retrained the 2D keypoint detector on the HO3D v2 dataset. Additionally, my work is from two years ago, and it's not straightforward to replicate it end-to-end. Moreover, metrics without procrustes alignment to be poor, so I recommend following recent work instead. image

llllljw commented 7 months ago

Sorry to bother you again. I have trained the 2D keypoints detector again on HO3D_v2 again, and I have visualize the 2D keypoints results, it works. And I use this as kypt_eval.json, but the results are still almost the same as past. ALUEQ1 V 0{E2V5T(M9_7ND

heartStrive commented 7 months ago

Where did you download the HO3D dataset? Suggested download from HandOccNet.

llllljw commented 7 months ago

Thanks for your response. I have download the HO3D dataset from the link that project:HandOccNet have provided:

image
llllljw commented 7 months ago

And it's weird that the 20th epoch is better than the 70th epoch }BJ$)S8 ZQ0`}I1@JDR07KE

wang-yan-adrian commented 5 months ago

I would like to ask, where is the code for calculating evaluation indicators

wang-yan-adrian commented 5 months ago

And it's weird that the 20th epoch is better than the 70th epoch }BJ$)S8 ZQ0`}I1@JDR07KE

I would like to ask, where is the code for calculating evaluation indicators

YiJian666 commented 3 months ago

你在哪里下载了 HO3D 数据集？建议从 HandOccNet 下载。

您好，我一直关注您的工作，请问您在HandOCCnet下提出的问题，最后是怎么解决的呢，希望您能为我解惑，不胜感激

7279eaec5c37f881aff5832016de6532