In this post, I explain transformation equivariance (TE) loss which is an unsupervised constraint for learning pixel-level labels in segmentation, keypoint localization or parts detection tasks.
Apr 25, 2021
