Research map / Architectures and efficiency / Pruning and efficient tuning
Parameter-Free Fine-tuning via Redundancy Elimination for Vision Foundation Models
Jiahuan Long, Tingsong Jiang, Wen Yao, Yizhe Xiong, Zhengqin Xu, Shuai Jia, Hanqing Liu, Chao Ma
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 |
|---|---|---|---|
| replace redundant channels with effective channels | SAM ViT-B · 9 segmentation datasets (avg) | mIoU: 49.14 -> 58.08 | Improved |
| replace redundant channels with effective channels | SAM ViT-H · 9 segmentation datasets (avg) | mIoU: 55.54 -> 60.68 | Improved |
| replace redundant channels with effective channels | SAM2 Hiera-L · 9 segmentation datasets (avg) | mIoU: 67.77 -> 73.53 | Improved |
| replace redundant channels with effective channels on in-domain VOC | SAM ViT-B · VOC | mIoU: 38.88 -> 56.33 | Improved |
| replace redundant channels on out-of-domain ISIC16 | SAM ViT-B · ISIC16 | mIoU: 57.33 -> 65.21 | Improved |
| replace redundant channels on top of SAM-PARSER | SAM-PARSER · 9 segmentation datasets (avg) | mIoU: 60.96 -> 65.39 | Improved |
| replace redundant channels on top of DoRA | DoRA · 9 segmentation datasets (avg) | mIoU: 79.12 -> 79.92 | Improved |
| replace redundant channels on top of encoder-only tuning | Encoder-only tuning · COD10K | mIoU: 50.65 -> no gain ('-') | No clear effect |
| replace redundant channels on DINOv2 ViT-S for classification | DINOv2 ViT-S · CIFAR | accuracy (%): 80.41 -> 80.81 | Improved |
| replace redundant channels on DINOv2 ViT-B for depth | DINOv2 ViT-B · NYUv2 | AbsRel: 0.110 -> 0.095 | Improved |
| zero out single channels of SAM (ablation) | SAM · PerSeg | mIoU: 50.6 -> 52.7 (ch 216), 47.9 (ch 189) | Mixed |
| increase number of replacement pairs (ablation) | channel replacement · COCO/CAMO/KVASIR (Hiera-S/T) | mIoU | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- SAM-PARSER — SAM-PARSER: Fine-Tuning SAM Efficiently by Parameter Space Reconstruction (AAAI 2024) · better on 2 of 2 shared results
- DoRA — DoRA: Weight-Decomposed Low-Rank Adaptation (ICML 2024) · 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
- Parameter Efficient Fine-tuning via Cross Block Orchestration for Segment Anything Model (CVPR 2024 · similarity 0.88)
- Low-Rank Rescaled Vision Transformer Fine-Tuning: A Residual Design Approach (CVPR 2024 · similarity 0.87)
- Minimal Interaction Seperated Tuning: A New Paradigm for Visual Adaptation (CVPR 2025 · similarity 0.87)
- Double-Filter: Efficient Fine-tuning of Pre-trained Vision-Language Models via Patch&Layer Filtering (ICML 2025 · similarity 0.87)
- VFM-Adapter: Adapting Visual Foundation Models for Dense Prediction with Dynamic Hybrid Operation Mapping (AAAI 2025 · similarity 0.87)
- 5%>100%: Breaking Performance Shackles of Full Fine-Tuning on Visual Recognition Tasks (CVPR 2025 · similarity 0.87)
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.
