Research map / Architectures and efficiency / Pruning and efficient tuning
Dual-Kernel Adapter: Expanding Spatial Horizons for Data-Constrained Medical Image Analysis
Ziquan Zhu, Hanruo Zhu, Si-Yuan Lu, Xiang Li, Yanda Meng, Yunxiao Zhang, Gaojie Jin, Lu Yin, Lijie Hu, Di Wang, Lu Liu, Tianjin Huang
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 standard Adapter instead of linear probing when training data is 0.63% (medical) | Adapter · BUSI classification (0.63%) | ACC (%): 73.48 -> 63.18 | Hurt |
| use standard Adapter instead of linear probing at 0.63% COVID | Adapter · COVID (0.63%) | ACC (%): 86.84 -> 83.29 | Hurt |
| use DKA instead of standard Adapter | DKA · BUSI classification (0.63%) | ACC (%): 63.18 -> 74.23 | Improved |
| use DKA instead of Adapter at full data | DKA · COVID (100%) | ACC (%): 98.33 -> 99.21 | Improved |
| use DKA instead of Adapter with RadImageNet ResNet-50 | DKA · ISIC-2019 (0.63%) | ACC (%): 51.32 -> 53.69 | Improved |
| use DKA instead of Adapter with MedSAM | DKA · BUSI segmentation (0.63%) | mIoU (%): 35.40 -> 37.13 | Improved |
| increase middle dimension from 1 to 16 (ablation) | DKA · BUSI classification (0.63%) | ACC (%): 65.02 -> 74.23 | Improved |
| increase middle dimension from 16 to 32 (ablation) | DKA · BUSI classification (0.63%) | ACC (%): 74.23 -> 73.75 | No clear effect |
| increase segmentation middle dimension from 64 to 192 (ablation) | DKA · ISIC-2018 (0.63%) | mIoU (%): 57.93 -> 63.13 | Improved |
| replace fixed dual kernels with learnable kernel selection (ablation) | DKA · COVID (0.63%) | ACC (%): 89.01 -> 88.98 | No clear effect |
| use dual-branch (5x5 + 51x51) instead of single branch (ablation) | DKA · COVID, BUSI, ISIC-2019 | ACC (%) | Improved |
| use asynchronous learning rates for adapter and head (1e-3 vs 1e-4) (ablation) | DKA · COVID (0.63%, 1.25%) | ACC (%) | Improved |
Closest papers by meaning
- FOCUS: Benchmarking Retinal Model Generalization from Foundation Vision Encoders to Multimodal LLMs (NeurIPS 2026 · similarity 0.85)
- Unleashing Foundation Vision Models: Adaptive Transfer for Diverse Data-Limited Scientific Domains (NeurIPS 2025 · similarity 0.85)
- Perturbating, Tuning, and Collaborating: Harnessing Vision Foundation Models for Single Domain Generalization on Medical Imaging (AAAI 2025 · similarity 0.84)
- ProKeR: A Kernel Perspective on Few-Shot Adaptation of Large Vision-Language Models (CVPR 2025 · similarity 0.84)
- TinyMIG: Transferring Generalization from Vision Foundation Models to Single-Domain Medical Imaging (ICML 2025 · similarity 0.84)
- VFM-Adapter: Adapting Visual Foundation Models for Dense Prediction with Dynamic Hybrid Operation Mapping (AAAI 2025 · 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.
