aofrancani/TSformer-VO

06 sequence

Closed this issue · 2 comments

Why is the 06 sequence result so poor? May I ask if there are any improvement measures?

My model struggled to estimate rotation values, I believe this is the reason for this poor result in sequence 6. Computing a weighted loss for rotation and translation separately could help, giving a higher weight for rotation as they did in the DeepVO model. However, you'll have to tune another hyperparameter during your experiments to find this weight.

Thank you very much for your reply. I will try it out.