├── img └── Teaser.png └── README.md /img/Teaser.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Jacky1128/Rep-MTL/HEAD/img/Teaser.png -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 |
2 | 3 | # Rep-MTL: Unleashing the Power of Representation-level Task Saliency for Multi-Task Learning 4 | 5 | 6 |
7 | 8 |
9 | 10 |

11 | 12 | arXiv 13 | 14 | 15 | Project Page 16 | 17 | 18 | HuggingFace Daily Top 5 19 | 20 |

21 | 22 | 23 | **Zedong Wang1, Siyuan Li2, Dan Xu1** 24 | 25 | 1The Hong Kong University of Science and Technology, 2Zhejiang University 26 | 27 |
28 | 29 |
30 |
31 | 32 | 33 |
34 | Rep-MTL Method Overview 35 |

(a) Most existing MTL optimization methods focus on addressing conflicts in parameter updates. (b) Rep-MTL instead leverages task saliency in shared representation space to explicitly facilitate cross-task information sharing while preserving task-specific signals via regularization, without modifications to either the underlying optimizers or model architectures.

36 |
37 | 38 | 39 | 40 | ## Overview 41 | 42 | Rep-MTL is a representation-level regularization method for multi-task learning that introduces task saliency-based objectives to encourage inter-task complementarity via Cross-task Saliency Alignment (CSA) while mitigating negative transfer among tasks via Task-specific Saliency Regulation (TSR). 43 | 44 | 45 | 46 | ## Benchmarks 47 | 48 | We evaluate Rep-MTL on several challenging MTL benchmarks spanning diverse computer vision scenarios: 49 | 50 | | Dataset | Tasks | Scenario | Download | 51 | |---------|-------|--------|----------| 52 | | NYUv2 | SemSeg + Depth Est. + Surface Normal Pred. | Indoor Dense Prediction | Link | 53 | | CityScapes | SemSeg + Depth Est. | Outdoor Dense Prediction | Link | 54 | | Office-31 | Image Classification (31 classes) | Domain Adaptation | Link | 55 | | Office-Home | Image Classification (65 classes) | Domain Adaptation | Link | 56 | 57 | 58 | 59 | ## Updates 60 | - **[July 24, 2025]** 🎉 Rep-MTL was selected as **ICCV 2025 Highlight**! We are working on cleaning and organizing our codebase. Stay tuned! 61 | - **[June 26, 2025]** Rep-MTL was accepted to **ICCV 2025**, with final ratings: 5/5/6 (out of 6). 62 | 63 | 64 | ## Contact 65 | 66 | For questions or research discussions, please contact Zedong Wang at `zedong.wang@connect.ust.hk`. 67 | 68 | 69 | 70 | ## BibTeX 71 | 72 | 73 | ```bibtex 74 | @inproceedings{iccv2025repmtl, 75 | title={Rep-MTL: Unleashing the Power of Representation-level Task Saliency for Multi-Task Learning}, 76 | author={Wang, Zedong and Li, Siyuan and Xu, Dan}, 77 | booktitle={IEEE/CVF International Conference on Computer Vision (ICCV)}, 78 | year={2025} 79 | } 80 | ``` 81 | 82 | 83 | ## Acknowledgements 84 | 85 | We thank the following great repositories that facilitated our research: LibMTL, CAGrad, MTAN, FAMO, and Nash-MTL. We also extend our appreciation to many other studies in the community for their foundational contributions that inspired this work. 86 | --------------------------------------------------------------------------------