Research map / Architectures and efficiency / Pruning and efficient tuning
Polyhistor: Parameter-Efficient Multi-Task Adaptation for Dense Vision Tasks
Yen-Cheng Liu, Chih-Yao Ma, Junjiao Tian, Zijian He, Zsolt Kira
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 decomposed low-rank hypernetworks (Polyhistor vs Hyperformer) | Hyperformer · PASCAL-Context | avg relative improvement (%): 2.64 -> 2.34 with 72.77M -> 6.41M encoder params | No clear effect |
| add layer-wise scaling kernels and cross-layer hypernetwork sharing (Polyhistor-Lite) (ablation) | Polyhistor · PASCAL-Context | avg relative improvement (%): 2.34 -> 1.74 with 6.41M -> 0.41M encoder params | Mixed |
| increase task embedding size from 16 to 64 (ablation) | Polyhistor-Lite · PASCAL-Context | avg relative improvement (%): -0.15 -> 1.74 | Improved |
| increase hypernetwork output rank from 1 to n/2 (ablation) | Polyhistor · PASCAL-Context | avg relative improvement (%): 0.52 -> 2.84 | Improved |
| shrink Hyperformer hypernetwork (rho 8 to 32) (ablation) | Hyperformer · PASCAL-Context | avg relative improvement (%): 2.64 -> -0.14 | Hurt |
| use larger backbone Swin-B instead of Swin-T (both IN-1k) (ablation) | Polyhistor-Lite · PASCAL-Context | avg relative improvement (%): 1.74 -> 6.38 (rho=1 on Swin-B) | Improved |
| pretrain backbone on ImageNet-22k instead of 1k (ablation) | Polyhistor-Lite · PASCAL-Context | avg relative improvement (%): 6.38 -> 8.08 (rho=1) | Improved |
| tune adapter hidden size of baselines (Adapter, Compacter, Shared Adapter) (ablation) | baseline adapters · PASCAL-Context | avg relative improvement (%): -2.71 -> -3.03 (Adapter rho 2 -> 4) | No clear effect |
| VPT input-space prompts vs decoders only | VPT-deep · PASCAL-Context | avg relative improvement (%): -11.02 -> -10.85 | No clear effect |
Later papers that compared with this method
2 later accepted papers report Polyhistor as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Parameter-, Memory-, Time-Efficient Multi-Task Dense Vision Adaptation (AAAI 2026 · better on 1 of 1 shared result)
- VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 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
- VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 2024 · similarity 0.89)
- TADFormer: Task-Adaptive Dynamic TransFormer for Efficient Multi-Task Learning (CVPR 2025 · similarity 0.87)
- Rethinking Few-Shot Adaptation of Vision-Language Models in Two Stages (CVPR 2025 · similarity 0.86)
- What Matters When Repurposing Diffusion Models for General Dense Perception Tasks? (ICLR 2025 · similarity 0.86)
- p-Laplacian Adaptation for Generative Pre-trained Vision-Language Models (AAAI 2024 · similarity 0.86)
- Connecting the Dots: Collaborative Fine-tuning for Black-Box Vision-Language Models (ICML 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.
