Skip to content

Questions about training procedure #5

Description

@BrianWayland

@Uehwan Thank you for sharing your implementation kindly. When learning your code I find two problems:

  • in train.py, the dataset is didn't looped through and just used the first batch for training, as shown in follows:
   ds = iter(trainer.train_loader)
   inputs = next(ds)
  • after predict depth and pose, the loss is not calculated to back propagate gradient and optimize parameters.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions