Evaluation for occluded/non-occluded points

[hurjunhwa]

Hello, Thanks for the great work and source codes!

I have a short question regarding the evaluation for occluded/non-occluded points.

According to the data processing code,

https://github.com/laoreja/HPLFlowNet/blob/master/data_preprocess/process_flyingthings3d_subset.py https://github.com/laoreja/HPLFlowNet/blob/master/data_preprocess/process_kitti.py

It collects non-occluded points for FlyingThings3D dataset and all valid points for KITTI dataset. Then for the scene flow evaluation, does it mean that it evaluates only non-occluded points for FlyingThings3D and all valid points for KITTI?

Best, Jun

[BlarkLee]

Same question

[laoreja]

Hi,

Thanks for the question!

The research was done several years ago, so I may not remember all the details.

FlyingThings3D is synthetic while KITTI is realistic. For realistic dataset, I feel there will not be ground truth for occluded points.

Also, KITTI’s ground truth is sparse.

You can read more about KITTI on their official materials.

On Wed, Aug 25, 2021 at 5:23 PM Runfa(Blark) Li @.***> wrote:

Same question

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/laoreja/HPLFlowNet/issues/25#issuecomment-905963125, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACHPMN2YENZ4B2LQSADL323T6WCPBANCNFSM45CL3VJQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email .

-- Sent from my iPhone

[hurjunhwa]

Thanks for your reply! KITTI actually has GT for occluded points as well.

So, I can conclude that the code evaluates non-occluded points on FlyingThings3D and all (sparse) points on KITTI.

Have a nice day!