Research map / Label-efficient and robust learning / Recognition and long-tail
Contrastive Multi-Task Dense Prediction
Siwei Yang, Hanrong Ye, Dan Xu
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 |
|---|---|---|---|
| Apply CCR to InvPT | InvPT · NYUD-v2 | SemSeg mIoU: 52.84 -> 53.80 | Improved |
| Apply CCR to InvPT | InvPT · Pascal-Context | Normal mErr: 14.15 -> 13.93 | Improved |
| Add basic cross-task contrastive regularization (ablation) | MT Baseline · NYUD-v2 | SemSeg mIoU: 38.901 -> 39.499 | Improved |
| Add a shared feature projector (ablation) | CCR-Basic · NYUD-v2 | SemSeg mIoU: 39.499 -> 39.956 | Improved |
| Add semi-hard pixel sampling (ablation) | CCR + Proj · NYUD-v2 | SemSeg mIoU: 39.956 -> 40.779 | Improved |
| Add contrastive task-pair selection (ablation) | CCR + Proj + SS · NYUD-v2 | SemSeg mIoU: 40.779 -> 41.275 | Improved |
| Share projectors among source tasks rather than leaving projectors unshared (ablation) | CCR with SS and CTS · NYUD-v2 | SemSeg mIoU: 40.414 -> 41.275 | Improved |
| Share projectors among target tasks (ablation) | CCR with SS and CTS · NYUD-v2 | SemSeg mIoU: 40.414 -> 40.273 | Hurt |
| Set the top-k factor to 128 (ablation) | CCR · NYUD-v2 | Overall multi-task performance | Improved |
| Increase the sampling ratio γ (ablation) | CCR · NYUD-v2 | Overall multi-task performance | Improved |
| Add CCR | PAD-Net · NYUD-v2 | SemSeg mIoU: 35.406 -> 36.925 | Improved |
| Add CCR | MTI-Net · NYUD-v2 | SemSeg mIoU: 37.456 -> 39.192 | Improved |
Closest papers by meaning
- Multi-Task Dense Prediction via Mixture of Low-Rank Experts (CVPR 2024 · similarity 0.86)
- Going Beyond Multi-Task Dense Prediction with Synergy Embedding Models (CVPR 2024 · similarity 0.86)
- Joint-Task Regularization for Partially Labeled Multi-Task Learning (CVPR 2024 · similarity 0.85)
- Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (ICML 2024 · similarity 0.84)
- Parameter-Efficient Multi-Task Model Fusion with Partial Linearization (ICLR 2024 · similarity 0.84)
- DiffusionMTL: Learning Multi-Task Denoising Diffusion Model from Partially Annotated Data (CVPR 2024 · 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.
