Research map / Detection and segmentation / Open-set and few-shot detection
ContrastMask: Contrastive Learning To Segment Every Thing
Check 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 |
|---|---|---|---|
| Use ContrastMask instead of CPMask | COCO partially-supervised instance segmentation with ResNet-101, 3× · COCO val2017 nonvoc→voc | mAP: 36.8 -> 38.4 | Helped |
| Use ContrastMask instead of CPMask | COCO partially-supervised instance segmentation with ResNet-101, 3× · COCO val2017 voc→nonvoc | mAP: 34.0 -> 34.3 | Helped |
| Add CAM to the Mask R-CNN mask head (ablation) | Mask R-CNN · COCO val2017 nonvoc→voc | mAP: 23.9 -> 32.3 | Helped |
| Add the CL Head to the CAM-augmented baseline (ablation) | Mask R-CNN + CAM · COCO val2017 nonvoc→voc | mAP: 32.3 -> 35.1 | Helped |
| Increase the CL encoder from 4 to 8 Conv-ReLU blocks and use a 3-layer projector (ablation) | ContrastMask CL Head · COCO val2017 nonvoc→voc | mAP: 34.2 -> 35.1 | Helped |
| Increase the CL encoder to 12 Conv-ReLU blocks and projector to 4 layers (ablation) | ContrastMask CL Head · COCO val2017 nonvoc→voc | mAP: 35.1 -> 35.0 | No clear effect |
| Set sampling ratio σ to 0.1 instead of 0.3 (ablation) | ContrastMask · COCO val2017 nonvoc→voc | mAP: 35.1 -> 34.4 | Hurt |
| Set sampling ratio σ to 0.2 instead of 0.3 (ablation) | ContrastMask · COCO val2017 nonvoc→voc | mAP: 35.1 -> 34.7 | Hurt |
| Set sampling ratio σ to 0.6 instead of 0.3 (ablation) | ContrastMask · COCO val2017 nonvoc→voc | mAP: 35.1 -> 34.3 | Hurt |
| Set easy-key temperature τ to 0.1 instead of 0.7 (ablation) | ContrastMask · COCO val2017 nonvoc→voc | mAP: 35.1 -> 34.4 | Hurt |
| Set easy-key temperature τ to 0.9 instead of 0.7 (ablation) | ContrastMask · COCO val2017 nonvoc→voc | mAP: 35.1 -> 34.0 | Hurt |
| Use contrastive supervision from base and novel categories rather than base categories alone (ablation) | ContrastMask · COCO val2017 nonvoc→voc | mAP: 33.5 -> 35.1 | Helped |
Closest papers by meaning
- Learning to Segment Every Thing (CVPR 2018 · similarity 0.89)
- MaskDiff: Modeling Mask Distribution with Diffusion Probabilistic Model for Few-Shot Instance Segmentation (AAAI 2024 · similarity 0.89)
- BoxInst: High-Performance Instance Segmentation With Box Annotations (CVPR 2021 · similarity 0.89)
- Revisiting Contrastive Methods for Unsupervised Learning of Visual Representations (NeurIPS 2021 · similarity 0.88)
- DynaMask: Dynamic Mask Selection for Instance Segmentation (CVPR 2023 · similarity 0.87)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · 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.