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

nucenes-mini Dataset: depth trainning problems #25

Open
JLeere opened this issue Aug 5, 2024 · 1 comment
Open

nucenes-mini Dataset: depth trainning problems #25

JLeere opened this issue Aug 5, 2024 · 1 comment

Comments

@JLeere
Copy link

JLeere commented Aug 5, 2024

image
您好, 非常感谢你出色的工作,, 我在复现模型时遇到一些问题请教一下. 我使用nucenes-mini 数据集在3090中微调depth模型, 我将Voxel-size参数调到您推荐的大小, 然后将contracted_coord, infinite_range,auxiliary_frame都设置为了False, 打印出generate_images_pred预测的图像很多都是大面积的失真, 然后观察输出的深度的min和max差距很小, 均为最小0.1附近.
按照您的经验, 这可能是什么原因造成的呢?

@LinShan-Bin
Copy link
Owner

感谢您的关注!有三个可能:

  1. 改变超参下微调模型时,预训练模型可能不是一个很好的初始化;
  2. 模型刚开始训练的时候初始深度较小,投影时全部被 mask 掉而没有投到参考图片上,可以尝试增大采样的最小距离;
  3. loss_distortion 的权重过大,可能要根据其他参数调整,尝试降低 loss_distortion 的权重。

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