Skip to content

overfit to context views #89

@LiuChunlin-eng

Description

@LiuChunlin-eng

Sorry to bother you again. I tried to go through the training process on the dl3dv dataset. When training on the dl3dv dataset, overfitting to the top and bottom views is not obvious or common. However, when I switch to a custom dataset and train for fewer steps, I notice the views are as follows:
Image
I want to ask:

  1. Is this phenomenon normal?
  2. My next step is to try loading the anysplat weights or modifying the data loading sampling strategy. Do you have any good suggestions?

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions