Research map / Detection and segmentation / Weakly supervised detection and segmentation
DaTaSeg: Taming a Universal Multi-Dataset Multi-Task Segmentation Model
Xiuye Gu, Yin Cui, Jonathan Huang, Abdullah Rashwan, Xuan Yang, Xingyi Zhou, Golnaz Ghiasi, Weicheng Kuo, Huizhong Chen, Liang-Chieh Chen, David A Ross
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 |
|---|---|---|---|
| Co-train across ADE semantic, COCO panoptic, and Objects365 detection instead of training a separate model per dataset | DaTaSeg with ResNet50 · ADE semantic | mIoU: 42.0 -> 48.1 (+6.1) | Improved |
| Co-train across ADE semantic, COCO panoptic, and Objects365 detection instead of training a separate model per dataset | DaTaSeg with ResNet50 · COCO panoptic | PQ: 48.2 -> 49.0 (+0.8) | Improved |
| Transfer knowledge from COCO panoptic while training ADE semantic supervision | DaTaSeg with ResNet50 · ADE semantic to panoptic | PQ: 26.9 -> 29.8 (+2.9) | Improved |
| Transfer instance-segmentation knowledge from COCO panoptic during box-supervised training | DaTaSeg with ResNet50 · Objects365 box to instance | mask AP: 12.3 -> 14.3 (+2.0) | Improved |
| Increase segment queries from 100 to 150 (ablation) | DaTaSeg with ResNet50 · ADE semantic | mIoU: 48.1 -> 48.7 | Improved |
| Increase segment queries from 100 to 150 (ablation) | DaTaSeg with ResNet50 · COCO panoptic | PQ: 49.0 -> 48.9 | No clear effect |
| Increase segment queries from 100 to 150 (ablation) | DaTaSeg with ResNet50 · ADE semantic to panoptic | PQ: 29.8 -> 29.1 | Hurt |
| Increase segment queries from 100 to 150 (ablation) | DaTaSeg with ResNet50 · Objects365 box to instance | AP: 14.3 -> 15.8 | Improved |
| Add dataset-specific queries, classification heads, and background classifiers (ablation) | DaTaSeg with ResNet50 · COCO panoptic | PQ: 49.0 -> 46.0 (-3.0) | Hurt |
| Add dataset-specific queries, classification heads, and background classifiers (ablation) | DaTaSeg with ResNet50 · ADE semantic to panoptic | PQ: 29.8 -> 26.9 (-2.9) | Hurt |
| Add dataset-specific queries, classification heads, and background classifiers (ablation) | DaTaSeg with ResNet50 · Objects365 box to instance | AP: 14.3 -> 10.9 (-3.4) | Hurt |
| Use 30 dataset-specific queries rather than none (ablation) | DaTaSeg with ResNet50 · COCO panoptic | PQ: 48.1 -> 45.8 | Hurt |
Closest papers by meaning
- OMG-Seg: Is One Model Good Enough For All Segmentation? (CVPR 2024 · similarity 0.87)
- Seg-ReSearch: Segmentation with Interleaved Reasoning and External Search (ICML 2026 · similarity 0.86)
- Exploring Simple Open-Vocabulary Semantic Segmentation (CVPR 2025 · similarity 0.86)
- Unsupervised Universal Image Segmentation (CVPR 2024 · similarity 0.86)
- Improving Semi-Supervised Semantic Segmentation with Sliced-Wasserstein Feature Alignment and Uniformity (CVPR 2025 · similarity 0.86)
- HyperSeg: Hybrid Segmentation Assistant with Fine-grained Visual Perceiver (CVPR 2025 · 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.
