Following your steps, I uploaded the trained parallax map to the kitti15 official website. The D1-fg (3.71) obtained is much higher than the 3.16 in your paper, but the D1-bg is the same as in the paper. What is the reason? thanks
hello, the uploaded checkpoint is for kitti12. We don't upload the kitti2015 checkpoint. you should finetune the model on the kitti 2015 dataset as we said in the paper
Following your steps, I uploaded the trained parallax map to the kitti15 official website. The D1-fg (3.71) obtained is much higher than the 3.16 in your paper, but the D1-bg is the same as in the paper. What is the reason? thanks