Research map / Image generation / Text-to-image diffusion
TaxoMIL: Taxonomy-Constrained Learning for Hierarchical Whole Slide Image Analysis
Chaeyeon Lee, Khang Quoc, Jinsol Song, Yosep Chong, Kwangil Yim, JIN TAE KWAK
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 |
|---|---|---|---|
| add cyclic loss scheduling to HIER, ITAL and CL (ablation) | TaxoMIL · GastWSI, BRACS, PANDA | holistic accuracy (%): GastWSI 63.07 -> 64.55; BRACS 72.22 -> 75.78; PANDA 51.41 -> 54.60 | Improved |
| add HIER and ITAL losses on top of GEN+CL (ablation) | TaxoMIL · BRACS | holistic accuracy (%): 62.96 -> 72.22 | Improved |
| add HIER and ITAL losses on top of GEN+CL (no cyclic) (ablation) | TaxoMIL · PANDA | holistic accuracy (%): 51.97 -> 51.41 | No clear effect |
| add supervised contrastive loss to GEN (ablation) | TaxoMIL · BRACS | holistic accuracy (%): 66.67 -> 62.96 | Hurt |
| add supervised contrastive loss to GEN (ablation) | TaxoMIL · GastWSI, PANDA | holistic accuracy (%): GastWSI 61.58 -> 62.60; PANDA 49.53 -> 51.97 | Improved |
| add label-context token e_ctx to decoder prefix (ablation) | TaxoMIL · GastWSI, BRACS, PANDA | holistic accuracy (%): GastWSI 62.51 -> 64.55; BRACS 70.37 -> 75.78; PANDA 50.56 -> 54.60 | Improved |
| add learnable image token <|img|> to decoder prefix (ablation) | TaxoMIL · GastWSI, BRACS, PANDA | holistic accuracy (%): GastWSI 62.05 -> 64.55; BRACS 70.37 -> 75.78; PANDA 53.10 -> 54.60 | Improved |
| use autoregressive text decoder instead of classification heads (ablation) | TaxoMIL · GastWSI, BRACS, PANDA | holistic accuracy (%): GastWSI 63.81 -> 64.55; BRACS 72.22 -> 75.78; PANDA 52.35 -> 54.60 | Improved |
| use GPT-2 small instead of GPT-2 medium decoder (ablation) | TaxoMIL · BRACS | holistic accuracy (%): 66.67 -> 75.78 | Improved |
| TaxoMIL vs strongest baseline per dataset (holistic) | TaxoMIL · GastWSI, BRACS, PANDA | holistic accuracy (%): GastWSI 63.26 -> 64.55; BRACS 68.52 -> 75.78; PANDA 50.28 -> 54.60 | Improved |
| flat MIL classifiers vs hierarchy-aware MIL (HMIL, Chang et al., HiClass) | hierarchical MIL baselines · PANDA, GastWSI | holistic accuracy (%): GastWSI HMIL 55.07 vs ABMIL 61.02 | No clear effect |
Closest papers by meaning
- CaMIL: Causal Multiple Instance Learning for Whole Slide Image Classification (AAAI 2024 · similarity 0.88)
- Generalizable Whole Slide Image Classification with Fine-Grained Visual-Semantic Interaction (CVPR 2024 · similarity 0.88)
- PathVQ: Reforming Computational Pathology Foundation Model for Whole Slide Image Analysis via Vector Quantization (NeurIPS 2025 · similarity 0.87)
- MLLM-HWSI: A Multimodal Large Language Model for Hierarchical Whole Slide Image Understanding (CVPR 2026 · similarity 0.87)
- Few-Shot Learning from Gigapixel Images via Hierarchical Vision-Language Alignment and Modeling (NeurIPS 2025 · similarity 0.87)
- ViLa-MIL: Dual-scale Vision-Language Multiple 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.
