Research map / Architectures and efficiency / Pruning and efficient tuning
AFFMAE: Scalable Vision Pre-Training for High-Resolution Microscopy Segmentation on Desktop Hardware
David Smerkous, Zian Wang, Behzad Najafian
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 AFFMAE instead of ViT-MAE for fine-tuning | high-res segmentation · FPW (1024px) | throughput img/s: 4 -> 20 | Improved |
| use AFFMAE instead of ViT-MAE for pre-training | MAE pre-training · Pre-training 512px | throughput img/s: 76 -> 151 | Improved |
| use AFFMAE instead of ViT-MAE | MAE pre-training · Pre-training 512px | peak memory GB: 29.7 -> 14.5 | Improved |
| use AFFMAE instead of ViT-MAE | FPW segmentation · FPW (mIoU, 512/768/1024) | mIoU: 0.606 -> 0.608 (512), 0.627 -> 0.622 (768), 0.630 -> 0.633 (1024) | No clear effect |
| use AFFMAE instead of ViT-MAE | FPW segmentation · FPW (1024px) | Slits IoU: .494 -> .514 | Improved |
| in-domain EM pretraining instead of ImageNet-1K init | MAE ViT · FPW 512px | mIoU: 0.494 -> 0.606 | Improved |
| add deep supervision (ablation) | AFFMAE 23M · FPW ablation (d_s=0.4) | mIoU: 0.5734 -> 0.5908 | Improved |
| reconstruct only from deepest stage instead of deep supervision (ablation) | AFFMAE 23M · FPW ablation | mIoU: 0.5908 -> 0.4353 | Hurt |
| increase token retention d_s from 0.25 to 0.5 (ablation) | AFFMAE 23M · FPW ablation | mIoU: 0.5729 -> 0.6009 | Improved |
| use d_s=0.4 rather than 0.5 (ablation) | AFFMAE 23M · FPW ablation | mIoU: 0.6009 -> 0.5908, GFLOPs 36 -> 30 | No clear effect |
| raise mask ratio from 0.50 to 0.75 (ablation) | AFFMAE 23M · FPW ablation | mIoU: 0.5908 -> 0.5739 | Hurt |
| Perlin masking instead of random masking (ablation) | AFFMAE 23M · FPW ablation (d_s=0.5) | mIoU: 0.5947 -> 0.6009 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SimMIM — SimMIM: A Simple Framework for Masked Image Modeling (CVPR 2022) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Structure-Guided Masked Autoencoders for Ultra-High Resolution Scientific Image Understanding (NeurIPS 2026 · similarity 0.86)
- ViTally Consistent: Scaling Biological Representation Learning for Cell Microscopy (ICML 2025 · similarity 0.86)
- Masked Autoencoders for Microscopy are Scalable Learners of Cellular Biology (CVPR 2024 · similarity 0.86)
- Micro-Bench: A Microscopy Benchmark for Vision-Language Understanding (NeurIPS 2024 · similarity 0.84)
- HMAR: Efficient Hierarchical Masked Auto-Regressive Image Generation (CVPR 2025 · similarity 0.84)
- Beyond ViT Tokens: Masked-Diffusion Pretrained Convolutional Pathology Foundation Model for Cell-Level Dense Prediction (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.
