Closed tyjiang1997 closed 3 years ago
Hi @LittleYuanzi, the val split is almost 50% of the total KITTI training data. Adding the val split to the training dataset should give you an edge on the testing results. Did you train the model using (train split + val split) and got worse test results than training with (train split) alone? What training configuration did you use? Thanks,
Hello, do you need to add the val dataset to the training to submit your results on the KITTI official website? I did so, but the accuracy decreased by 3 percentage points. Can you tell me the reason for this?