I have a question regarding the results on the paper (Table 1). Does these results imply that the model is trained without cropping the image into (192 x 640) ? So, if I want to reproduce the results on the paper, I have to change train_384rc.yaml and set the patch size to None ?
Hi,
Thanks for your attention!
The results in Table 1 of the paper were generated with the training settings in this repo, and random cropping was enabled in all the experiments.
Hello!
I have a question regarding the results on the paper (Table 1). Does these results imply that the model is trained without cropping the image into (192 x 640) ? So, if I want to reproduce the results on the paper, I have to change train_384rc.yaml and set the patch size to None ?
Thank you