Research map / Architectures and efficiency / Attention and transformers
MuViT: Multi-Resolution Vision Transformers for Learning Across Scales in Microscopy
Albert Dominguez Mantes, Gioele La Manno, Martin Weigert
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use true world-coordinate bounding boxes instead of naive centered ones (ablation) | MuViT[1,4]+UNETR · SYNTHETIC | mDSC: 0.3864 -> 0.9538 | Improved |
| add coarser resolution level (1 -> 1,4) (ablation) | MuViT+UNETR · SYNTHETIC | mDSC: 0.5003 -> 0.9538 | Improved |
| add resolution levels 8 and 32 to single-level encoder (ablation) | MuViT+Mask2Former · MOUSE (256x256 input) | mDSC: 0.391 -> 0.901 | Improved |
| use true world coordinates instead of naive (ablation) | MuViT[1,8,32]+Mask2Former · MOUSE | mDSC: 0.820 -> 0.901 | Improved |
| use true world coordinates instead of naive (ablation) | MuViT[1,8,32]+UNETR · MOUSE | mDSC: 0.793 -> 0.883 | Improved |
| add coarser level 8 to single-level encoder (ablation) | MuViT+UNETR · KPIS | DSC: 0.8492 -> 0.8958 | Improved |
| add coarser level 8 to single-level encoder (ablation) | MuViT+Mask2Former · KPIS | DSC: 0.8390 -> 0.8866 | Improved |
| add more resolution levels to the MAE encoder (1 -> 1,8 -> 1,8,32,64) (ablation) | MuViT-MAE frozen features · KPIS linear probing | OvA ROC-AUC: 0.958 -> 0.963 -> 0.988 | Improved |
| use true world coordinates instead of naive in MAE (ablation) | MuViT[1,8,32] MAE · MOUSE reconstruction | reconstruction MSE: level 8: 1.589 × 10−2 (naive) vs 2.274 × 10−2; level 32: 1.373 × 10−2 vs 1.464 × 10−2 | Mixed |
| use true world coordinates instead of naive in MAE (ablation) | MuViT[1,8,32,64] MAE · KPIS reconstruction | reconstruction MSE: level 64: 7.302 × 10−3 -> 6.913 × 10−3 | Improved |
| MAE pretraining of multi-resolution encoder | MuViT[1,8,32]+Mask2Former · MOUSE fine-tuning | validation mDSC: 0.843 after 10 epochs, above 0.88 by epoch 25 | Improved |
| increase input size for DeepLabV3 baseline from 256 to 1024 (ablation) | DeepLabV3 · MOUSE | mDSC: 0.520 -> 0.676 -> 0.843 | Improved |
Closest papers by meaning
- ViTally Consistent: Scaling Biological Representation Learning for Cell Microscopy (ICML 2025 · similarity 0.87)
- ChAda-ViT : Channel Adaptive Attention for Joint Representation Learning of Heterogeneous Microscopy Images (CVPR 2024 · similarity 0.87)
- Micro-Bench: A Microscopy Benchmark for Vision-Language Understanding (NeurIPS 2024 · similarity 0.86)
- Masked Autoencoders for Microscopy are Scalable Learners of Cellular Biology (CVPR 2024 · similarity 0.85)
- scSplit: Bringing Severity Cognizance to Image Decomposition in Fluorescence Microscopy (NeurIPS 2025 · similarity 0.84)
- MorphoHELM: A Comprehensive Benchmark for Evaluating Representations for Microscopy-Based Morphology Assays (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
