Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

standard dm的一些实现的疑问和请教 #11

Open
CBQ-1223 opened this issue Jun 8, 2024 · 2 comments
Open

standard dm的一些实现的疑问和请教 #11

CBQ-1223 opened this issue Jun 8, 2024 · 2 comments

Comments

@CBQ-1223
Copy link

CBQ-1223 commented Jun 8, 2024

image
大佬您论文中说的standard dm的实现,有几个具体细节想请教您,因为想复线一下:
1.您是在room上做的没在maptr上实验哈
2.我理解你说的就是训练时 standard dm的实现:相当于改变两个数值,sigma_guide = 1.0,mu_guide = 0.0,然后想请教下损失函数是没有变化吗,还是每个gt加噪的实例预测X0然后和原本这个gt的坐标算loss,因为GS-DM没有训练分类损失, standard dm也没有这部分吗
3.推理时的standard dm,按论文中说的是从标准高斯噪声采样,那么这是坐标,请问类别信息会预测吗,具体是怎么得到类别预测的呢
超级感谢大佬答疑解惑

@woodfrog
Copy link
Owner

woodfrog commented Jun 9, 2024

  1. 其实两个数据集都跑了,结论差不多,在正文ablation study里就只放了一种。
  2. 对,就是去掉了guidance network相关的东西,其他没有变化,loss就是和原本edm一样的loss。你说的分类损失指的是RoomFormer/MapTR里每个vertex是否存在的loss吧?这个standard DM也是基于proposal generator给出的room/vertex数量的,没有去同时处理离散和连续变量。
  3. 同上,类别信息是用的RoomFormer/MapTR的结果。

你再有别的问题的话直接email我吧

@CBQ-1223
Copy link
Author

好的,谢谢大佬!!!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants