Student Collaboration Improves Self-Supervised Learning: Dual-Loss Adaptive Masked Autoencoder for Multiplexed Immunofluorescence Brain Images Analysis. (arXiv:2205.05194v2 [cs.CV] UPDATED)

Self-supervised learning (SSL) leverages the underlying data structure to
generate supervisory signals for training deep networks. This approach offers a
practical solution for learning with multiplexed immunofluorescence brain
images where data are often more abundant than human expert annotations. SSL
algorithms based on contrastive learning and image reconstruction have
demonstrated impressive performances. Unfortunately, these methods were
designed and validated mostly on natural images rather than biomedical images.
A few recent works have applied SSL to analyzing cell images. However, none …

analysis arxiv autoencoder brain collaboration cv images learning loss masked autoencoder self-supervised learning supervised learning