Research map / Detection and segmentation / Open-set and few-shot detection
Compositor: Bottom-Up Clustering and Compositing for Robust Part and Object Segmentation
Ju He, Jieneng Chen, Ming-Xian Lin, Qihang Yu, Alan L. Yuille
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 |
|---|---|---|---|
| replace direct pixel-to-object prediction with bottom-up part clustering followed by part-to-object compositing | Compositor · PartImageNet val and Pascal-Part val | part and object mIoU: 60.34 -> 61.44; 70.21 -> 71.78; 47.61 -> 48.01; 72.69 -> 74.35 | Improved |
| use bottom-up part clustering and compositing | Compositor · Occluded-PartImageNet-v1 | part and object mIoU: 50.23 -> 54.63; 56.73 -> 63.79 | Improved |
| train parts and objects jointly with separate pixel-based branches rather than hierarchical interaction | Compositor · PartImageNet and Pascal-Part | part and object mIoU: PartImageNet ResNet-50: 61.44 vs 58.02 part, 71.78 vs 70.44 object; Pascal-Part ResNet-50: 48.01 vs 46.60 part, 74.35 vs | Hurt |
| set the part-to-object loss weight beta to 1/4 instead of 1/2 (ablation) | Compositor · Pascal-Part val | part and object mIoU: 55.92 -> 55.04; 83.10 -> 81.97 | Hurt |
| set the part-to-object loss weight beta to 1/3 instead of 1/2 (ablation) | Compositor · Pascal-Part val | part and object mIoU: 55.92 -> 55.36; 83.10 -> 82.65 | Hurt |
| increase beta to 1 instead of 1/2 (ablation) | Compositor · Pascal-Part val | part and object mIoU: 55.92 -> 53.38; 83.10 -> 82.03 | Hurt |
| increase beta to 2 instead of 1/2 (ablation) | Compositor · Pascal-Part val | part and object mIoU: 55.92 -> 51.41; 83.10 -> 81.74 | Hurt |
| increase query counts from 15/50 to 20/50 (ablation) | Compositor · Pascal-Part val | part and object mIoU: 54.68 -> 55.92; 82.03 -> 83.10 | Improved |
| increase part queries from 20 to 25 while keeping 50 object queries (ablation) | Compositor · Pascal-Part val | part and object mIoU: 55.92 -> 55.30; 83.10 -> 82.48 | Hurt |
| increase object queries from 50 to 60 while keeping 20 part queries (ablation) | Compositor · Pascal-Part val | part and object mIoU: 55.92 -> 54.71; 83.10 -> 82.21 | Hurt |
Later papers that compared with this method
2 later accepted papers report Compositor as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Knowledge-Guided Part Segmentation (ICCV 2025 · better on 4 of 4 shared results)
- From Pixels to Objects: A Hierarchical Approach for Part and Object Segmentation Using Local and Global Aggregation (ECCV 2024 · better on 6 of 6 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Fine-Grained Image-Text Correspondence with Cost Aggregation for Open-Vocabulary Part Segmentation (CVPR 2025 · similarity 0.88)
- From Pixels to Objects: A Hierarchical Approach for Part and Object Segmentation Using Local and Global Aggregation (ECCV 2024 · similarity 0.88)
- PartDistillation: Learning Parts From Instance Segmentation (CVPR 2023 · similarity 0.88)
- Towards Open-World Segmentation of Parts (CVPR 2023 · similarity 0.88)
- Task-aligned Part-aware Panoptic Segmentation through Joint Object-Part Representations (CVPR 2024 · similarity 0.86)
- Understanding Multi-Granularity for Open-Vocabulary Part Segmentation (NeurIPS 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.
