[2602.24222] MuViT: Multi-Resolution Vision Transformers for Learning Across Scales in Microscopy

Computer Science > Computer Vision and Pattern Recognition arXiv:2602.24222 (cs) [Submitted on 27 Feb 2026] Title:MuViT: Multi-Resolution Vision Transformers for Learning Across Scales in Microscopy Authors:Albert Dominguez Mantes, Gioele La Manno, Martin Weigert View a PDF of the paper titled MuViT: Multi-Resolution Vision Transformers for Learning Across Scales in Microscopy, by Albert Dominguez Mantes and 2 other authors View PDF HTML (experimental) Abstract:Modern microscopy routinely produces gigapixel images that contain structures across multiple spatial scales, from fine cellular morphology to broader tissue organization. Many analysis tasks require combining these scales, yet most vision models operate at a single resolution or derive multi-scale features from one view, limiting their ability to exploit the inherently multi-resolution nature of microscopy data. We introduce MuViT, a transformer architecture built to fuse true multi-resolution observations from the same underlying image. MuViT embeds all patches into a shared world-coordinate system and extends rotary positional embeddings to these coordinates, enabling attention to integrate wide-field context with high-resolution detail within a single encoder. Across synthetic benchmarks, kidney histopathology, and high-resolution mouse-brain microscopy, MuViT delivers consistent improvements over strong ViT and CNN baselines. Multi-resolution MAE pretraining further produces scale-consistent representations tha...