search

TRAILab / CaDDN

Categorical Depth Distribution Network for Monocular 3D Object Detection (CVPR 2021 Oral)
Apache License 2.0
359 stars 62 forks source link

About the waymo results. #83

Closed Cc-Hy closed 2 years ago

Cc-Hy commented 2 years ago

Hello, when doing the experiment on Waymo, you used the label in 'projected_lidar_labels[0]' of the original '.tfrecord‘ Waymo data files to calculate the AP, right? Or are the labels outside the point cloud range [2, 55.76] × [−25.6, 25.6] × [−4, 4] removed?

codyreading commented 2 years ago

Hello!

Are labels inside 'projected_lidar_labels[0] only the labels in the front view camera? If so, then that is correct.

For the record, our original Waymo implementation was with a Waymo dataset formatted like KITTI (see here). Which generated labels independently for each camera. I assume it was using projected_lidar_labels[0] but never checked.

Cc-Hy commented 2 years ago

There are 5 cameras in Waymo and the camera_0 corresponds to the front camera. And I found the lines producing labels in the adapter your mentioned is image which means extracting labels individually from projected_lidar_labels [0] to projected_lidar_labels [4].

codyreading commented 2 years ago

Okay great, going to close this issue now then