JUGGHM/PENet_ICRA2021

How to infer PENet for KITTI object task?

Senwang98 opened this issue · 3 comments

Hi, @JUGGHM
I am doing 3D detection with KITTI dataset.
I converted lidar(.bin) into sparse depth map(.png) now.
How can I get dense depth map using PENet?
(It seems that PENet support only one input shape, the kitti 3d detection dataset have different image shape)
Can you give me some advice? Thanks

Thanks for your interest! It is recommend that you could resize the kitti depth dataset into an expected image size during training and train the depth completion models from scratch. Maybe in the future we would update our method for supporting a variety of different solutions. However I am currently working on some other projects so it could be after a quite long period.

@JUGGHM
Thanks for your reply!
So, If I still using your pretrained model for test, I need to resize all my dataset into the same size?
Currently, I just want to using provided PENet model to see some results.(233333)

Jax29 commented

Hi, @JUGGHM I am doing 3D detection with KITTI dataset. I converted lidar(.bin) into sparse depth map(.png) now. How can I get dense depth map using PENet? (It seems that PENet support only one input shape, the kitti 3d detection dataset have different image shape) Can you give me some advice? Thanks

Bro, did you get the dense depth map? Can I talk to you?