Research map / Detection and segmentation / Open-set and few-shot detection
Multiscale Combinatorial Grouping for Image Segmentation and Object Proposal Generation
Jordi Pont-Tuset, Pablo Arbeláez, Jonathan T. Barron, Ferran Marqués, Jitendra Malik
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| go from single-scale to multiscale hierarchy (combined cues) | MCG segmenter · BSDS500 val | Fb ODS: 0.719 -> 0.725 | Improved |
| go from single-scale to multiscale hierarchy (combined cues) | MCG segmenter · BSDS500 val | SC ODS: 0.602 -> 0.611 | Improved |
| go from single-scale to multiscale hierarchy (combined cues) | MCG segmenter · BSDS500 val | VI ODS (lower is better): 1.619 -> 1.548 | Improved |
| go from single-scale to multiscale hierarchy (Pb input) | MCG segmenter · BSDS500 val | Fop ODS: 0.334 -> 0.350 | Improved |
| combine all cues instead of structured forests alone | single-scale segmenter · BSDS500 val | Fop ODS: 0.338 -> 0.358 | Improved |
| add downsampled normalized cuts (squaring and decimation) | contour globalization · BSDS500 val | speed: 20x speed-up | Improved |
| increase number of decimation steps d beyond 2 or 3 | DNCuts · BSDS500 | accuracy and speed | Hurt |
| add alignment with strength transfer to finest superpixels (vs resizing contours only) (ablation) | multiscale segmentation · BSDS500 val | segmentation quality: fails to improve over the single-scale result without it | Improved |
| add multiscale input to contour detector (vs original resolution only) (ablation) | multiscale segmentation · BSDS500 val | segmentation quality: small degradation when removed | Improved |
| use learned combiners (SVM, logistic regression, random forests) instead of uniform weights (ablation) | multiscale hierarchy · BSDS500 val | segmentation quality | No clear effect |
| use greedy Pareto front strategy instead of full combination (ablation) | proposal list combination · SegVOC12 train, reduced problem | Ji: mean loss 0.0002, max loss 0.004 | No clear effect |
| add Pareto optimization over equal distribution of proposals from four scales (ablation) | MCG proposals · SegVOC12 | Ji: gain of 2 points at working point | Improved |
Closest papers by meaning
- From Pixels to Objects: A Hierarchical Approach for Part and Object Segmentation Using Local and Global Aggregation (ECCV 2024 · similarity 0.87)
- Supervised Evaluation of Image Segmentation and Object Proposal Techniques (TPAMI 2016 · similarity 0.87)
- Unsupervised Hierarchical Semantic Segmentation With Multiview Cosegmentation and Clustering Transformers (CVPR 2022 · similarity 0.87)
- MultiSeg: Semantically Meaningful, Scale-Diverse Segmentations From Minimal User Input (ICCV 2019 · similarity 0.86)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.85)
- SPGroup3D: Superpoint Grouping Network for Indoor 3D Object Detection (AAAI 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.
