Research map / Detection and segmentation / Open-set and few-shot detection
Learning to Segment Every Thing
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 |
|---|---|---|---|
| Predict class-specific mask weights from box-detection weights instead of using class-agnostic masks | Mask R-CNN, ResNet-50-FPN · COCO val2017, voc → non-voc (B) | mask AP: 19.2 -> 23.7 | Helped |
| Predict class-specific mask weights from box-detection weights instead of using class-agnostic masks | Mask R-CNN, ResNet-50-FPN · COCO val2017, non-voc → voc (B) | mask AP: 23.9 -> 28.9 | Helped |
| Use visual box-classification and regression weights rather than random vectors as transfer input (ablation) | Weight transfer function · COCO val2017, voc → non-voc (B) | mask AP: 15.4 -> 20.2 | Helped |
| Use visual detection weights rather than GloVe vectors as transfer input (ablation) | Weight transfer function · COCO val2017, voc → non-voc (B) | mask AP: 17.3 -> 20.2 | Helped |
| Concatenate box-classification and box-regression weights as transfer input rather than using either alone (ablation) | Weight transfer function · COCO val2017, voc → non-voc (B) | mask AP: 18.1 -> 20.2; 19.8 -> 20.2 | Helped |
| Use a 2-layer LeakyReLU transfer function rather than the tested alternative structures (ablation) | Weight transfer function · COCO val2017, voc → non-voc (B) | mask AP: 19.2 -> 20.2 | Helped |
| Add a class-agnostic MLP mask branch to the class-agnostic baseline (ablation) | Mask R-CNN, ResNet-50-FPN · COCO val2017, voc → non-voc (B) | mask AP: 14.2 -> 17.1 | Helped |
| Add a class-agnostic MLP mask branch to weight transfer (ablation) | Mask R-CNN, ResNet-50-FPN · COCO val2017, voc → non-voc (B) | mask AP: 20.2 -> 21.3 | Helped |
| Train end-to-end rather than stage-wise (ablation) | Weight-transfer Mask R-CNN · COCO val2017, voc → non-voc (B) | mask AP: 20.2 -> 22.2 | Helped |
| Stop mask-loss gradients from updating box-detection weights during end-to-end training (ablation) | Weight-transfer Mask R-CNN · COCO val2017, voc → non-voc (B) | mask AP: 20.2 -> 22.2 | Helped |
| Use stage-wise rather than end-to-end training with stop-gradient (ablation) | Weight-transfer Mask R-CNN · COCO val2017, voc → non-voc (B) | mask AP: 22.2 -> 20.2 | Hurt |
| Train on a larger number of mask examples per class in the strongly supervised set | Weight-transfer Mask R-CNN · COCO random class splits, held-out B categories | mask AP relative increase: up to over 40% | Helped |
Closest papers by meaning
- BoxInst: High-Performance Instance Segmentation With Box Annotations (CVPR 2021 · similarity 0.92)
- Large-Scale Interactive Object Segmentation With Human Annotators (CVPR 2019 · similarity 0.90)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · similarity 0.89)
- Extreme Point Supervised Instance Segmentation (CVPR 2024 · similarity 0.88)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.88)
- Training-Free Dataset Pruning for Instance Segmentation (ICLR 2025 · similarity 0.88)
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.