Research map / Label-efficient and robust learning / Self-supervised and contrastive
Snuffy: Efficient Whole Slide Image Classifier
Hossein Jafarinia, Alireza Alipanah, Saeed Razavi, Nahal Mirzaie, Mohammad Hossein Rohban
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 |
|---|---|---|---|
| continual pre-training with Adapter instead of full fine-tuning (ablation) | Snuffy DINO · CAMELYON16 (slide-level) | AUC: 0.756 -> 0.936 | Improved |
| continual pre-training with Adapter instead of full fine-tuning (ablation) | Snuffy MAE · CAMELYON16 (slide-level) | AUC: 0.754 -> 0.910 | Improved |
| use Adapter continual pre-training instead of exhaustive DINO pre-training from scratch (ablation) | Snuffy DINO · CAMELYON16 (slide-level) | AUC: 0.987 -> 0.936 (time 15,393 -> 1,536 minutes) | Hurt |
| increase transformer depth from 1 to 5 layers (ablation) | Snuffy SimCLR Exhaustive · CAMELYON16 | slide metrics | Improved |
| increase random global attentions from 1 to 700 (ablation) | Snuffy SimCLR Exhaustive · CAMELYON16 | slide metrics | Improved |
| increase number of patches per WSI used in MAE adapter continual training (ablation) | Snuffy MAE Adapter · CAMELYON16 | slide metrics | Improved |
| replace attention-pooling MIL with Snuffy sparse transformer on MIL benchmarks | MIL pooling · MUSK2 | accuracy: ABMIL 0.812 vs Snuffy 0.789; Snuffy AUC 0.985 | Mixed |
| use Snuffy sparse attention instead of full ViT attention on natural images | Snuffy-S/16 · CIFAR-100 (from scratch) | top-1 accuracy: 0.535 -> 0.429 | Hurt |
| use Snuffy MIL pooling instead of DSMIL/ABMIL attention pooling | MIL pooling · CAMELYON16 (patch-level ROI) | FROC: 0.328 -> 0.622 vs DSMIL | 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.
- TransMIL — TransMIL: Transformer based Correlated Multiple Instance Learning for Whole Slide Image Classification (NeurIPS 2021) · 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
- DTFD-MIL: Double-Tier Feature Distillation Multiple Instance Learning for Histopathology Whole Slide Image Classification (CVPR 2022 · similarity 0.89)
- CaMIL: Causal Multiple Instance Learning for Whole Slide Image Classification (AAAI 2024 · similarity 0.88)
- Unsupervised Foundation Model-Agnostic Slide-Level Representation Learning (CVPR 2025 · similarity 0.87)
- Queryable Prototype Multiple Instance Learning with Vision-Language Models for Incremental Whole Slide Image Classification (AAAI 2025 · similarity 0.86)
- FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification (CVPR 2025 · similarity 0.86)
- Dynamic Policy-Driven Adaptive Multi-Instance Learning for Whole Slide Image Classification (CVPR 2024 · similarity 0.86)
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.
