Closed this issue 3 years ago · 1 comments
We need a new loss function that can handle and train on our root relative prediction data.
Work with camera relative coordinates instead and we dont have to fix this.