Occlusion Mask GT on KITTI dataset

[Magicboomliu]

Thank you for your outstanding work.

May I ask how can we obtain the Occlusion mask ground truth of KITTI dataset for supervision?
Thank you very much

[mli0603]

Hi @Magicboomliu

Thanks for your interest. The occlusion mask can be obtained by finding the disparities that are out of range (i.e., leads to negative image index after warpping to the right image) given the ground truth disparities.

You do not have to compute it yourself as it is handled automatically in the line below:

stereo-transformer/dataset/kitti.py

Line 98 in d0aa1ad

input_data = augment(input_data, self.transformation)

I hope it helps!

[Magicboomliu]

  您好，邮件已收到。感谢来信 ，我会尽快回复您。  Hello, I have already received the letter. Thanks for your mailing, I will reply as soon as possible.  こんにちは！確かご電子メールをもらいました。私はできるだけ早くあなたに返事を出します。