Research map / Label-efficient and robust learning / Recognition and long-tail
Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Learn task uncertainty to weight all three task losses instead of training semantic segmentation alone | Multi-Task Learning · Tiny CityScapes validation | IoU: 59.4% -> 63.4% | Helped |
| Learn task uncertainty to weight all three task losses instead of using equal loss weights | Multi-Task Learning · Tiny CityScapes validation | IoU: 50.1% -> 63.4% | Helped |
| Learn task uncertainty to weight all three task losses instead of approximate manually selected weights | Multi-Task Learning · Tiny CityScapes validation | Inverse depth mean error: 0.549 -> 0.522 | Helped |
| Add semantic segmentation to instance segmentation and depth uncertainty weighting (ablation) | 3-task uncertainty weighting · Tiny CityScapes validation | Inverse depth mean error: 0.539 -> 0.522 | Helped |
| Add depth regression to semantic and instance uncertainty weighting (ablation) | 3-task uncertainty weighting · Tiny CityScapes validation | IoU: 61.0% -> 63.4% | Helped |
| Add instance segmentation to semantic and depth uncertainty weighting (ablation) | 3-task uncertainty weighting · Tiny CityScapes validation | IoU: 62.7% -> 63.4% | Helped |
| Add semantic segmentation to instance and depth uncertainty weighting (ablation) | 3-task uncertainty weighting · Tiny CityScapes validation | Instance mean error: 3.54 -> 3.50 | Mixed |
Closest papers by meaning
- Taskology: Utilizing Task Relations at Scale (CVPR 2021 · similarity 0.87)
- Multi-Task Learning as Multi-Objective Optimization (NeurIPS 2018 · similarity 0.87)
- Multi-task Learning with 3D-Aware Regularization (ICLR 2024 · similarity 0.86)
- 3D-Aware Multi-Task Learning with Cross-View Correlations for Dense Scene Understanding (CVPR 2026 · similarity 0.85)
- DiTASK: Multi-Task Fine-Tuning with Diffeomorphic Transformations (CVPR 2025 · similarity 0.84)
- VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 2024 · similarity 0.83)
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.