Is there a way to save the augmented data within the training pipeline to enable future visualization?

[heming7]

I am training a 3D detection model on NuScenes dataset and I have applied data augmentation in the training pipeline. I would like to know is there a way that I can somehow save the augmented data so that I can use the render_pointcloud_in_image from the nuscenes-devkit to visualize the projection of lidar points onto images after data augmentation.

[ZCMax]

Sorry for the late reply, for now, we provide the browse_dataset.py script which can visualize the augmented data in the training pipeline, more details can be seen in https://github.com/open-mmlab/mmdetection3d/blob/master/tools/misc/browse_dataset.py. Because mmdet3d store point clouds and images in an object named DataContainer so you can take your point cloud and image after augment by: https://github.com/open-mmlab/mmdetection3d/blob/c8347b7ed933d70fcfbfb73a3541046b8c8e8f5e/tools/misc/browse_dataset.py#L110 https://github.com/open-mmlab/mmdetection3d/blob/c8347b7ed933d70fcfbfb73a3541046b8c8e8f5e/tools/misc/browse_dataset.py#L147 then continue to process convert torch to numpy and save it.