Search papers, labs, and topics across Lattice.
This paper introduces Mirror-Fusion-Augmented Self-Supervised Learning (MFASSL), a Vision Transformer framework that enhances self-supervised learning by incorporating a soft reflection prior to address the limitations of strict flip invariance in bilateral data. By constructing mirror-paired views and employing a Mirror-Fusion Attention (MFA) module, MFASSL allows for adaptive interactions between mirrored regions while maintaining asymmetric information. The framework demonstrates significant improvements in downstream tasks and robustness across multiple datasets, outperforming existing SSL methods with minimal additional parameters.
Mirror-Fusion Attention reveals that lightweight geometric priors can significantly enhance self-supervised learning performance without the need for extensive model redesigns.
Most self-supervised learning (SSL) methods encourage invariance across augmentations, but strict flip invariance can suppress informative left--right correspondences in approximately bilateral data such as medical images and human faces. We propose Mirror-Fusion-Augmented Self-Supervised Learning (MFASSL), a Vision Transformer framework that injects a soft reflection prior into standard SSL without redesigning the backbone. MFASSL constructs mirror-paired views aligned to an estimated symmetry axis and introduces a lightweight Mirror-Fusion Attention (MFA) module for adaptive token-level interaction between mirrored regions while preserving asymmetric cues. The base SSL objective is further coupled with reflection-consistency and mid-layer token-alignment losses. Across CheXpert, BraTS, CelebA-HQ, and WFLW, MFASSL improves downstream performance, calibration, and reflection robustness over MoCo-v3, DINO, and MAE baselines under matched ViT-B/16 settings. It also achieves stronger and more consistent gains than recent equivariant SSL approaches with only approximately 2.7\% additional parameters. These results show that lightweight geometry-aware priors can effectively complement invariance-based SSL.
