Where are the groundtruth poses used?

[jc3342]

Thanks for your work!
Maybe I missed something, but when I go through the code, in the data class DatasetVisibilityKittiSingle defined in DataVisibilityKitti.py, I didn't see that the groundtruth poses self.GTs_T and selfGTs_R are used. But if without groundtruth, how could you train the model?
Another question is, when doing inference, T_predicted, R_predicted = models[iteration](rgb, lidar) , error_t is norm of T_predicted? I thought it should be norm(T_predicted-T_groundtruth).
I guess I understand something wrong. But could you plz help me with that? Thanks a lot !

[cattaneod]

Thank you for your interest in our work.

The ground truths are embedded in the preprocessed point cloud, i.e., in the preprocessing step I save the local point cloud for each camera frame. This means that the ground truth for each (camera, point cloud) pair is the identity.

During training/inference, I apply a random transformation H_init to the point cloud, feed it to the network (together with the image), and the network predicts the transformation H_pred. The composed pose H_init*H_pred should be as close as possible to the identity transformation.

I hope my explanation is clear.