Research map / Detection and segmentation / CNN features
From Pixels to Objects: A Hierarchical Approach for Part and Object Segmentation Using Local and Global Aggregation
Yunfei Xie, Cihang Xie, Alan Yuille, Jieru Mei
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 bilinear upsampling with association-aware upsampling (ablation) | LGFormer · PartImageNet val | object mIoU: 73.2 -> 79.8 | Improved |
| replace bilinear upsampling with association-aware upsampling (ablation) | LGFormer · PartImageNet val | part mIoU: 65.3 -> 67.4 | Improved |
| use 64 group tokens instead of 256 (ablation) | LGFormer · PartImageNet val | object mIoU: 78.2 -> 79.8 | Improved |
| use 64 group tokens instead of 16 (ablation) | LGFormer · PartImageNet val | object mIoU: 78.6 -> 79.8 | Improved |
| initialize group tokens by average pooling instead of learnable tokens (ablation) | LGFormer · PartImageNet val | object mIoU: 79.3 -> 79.8 | Improved |
| use 3 GCA stages instead of 2 (ablation) | LGFormer · PartImageNet val | object mIoU: 77.7 -> 79.8 | Improved |
| use 3 GCA stages instead of 4 (ablation) | LGFormer · PartImageNet val | object mIoU: 78.5 -> 79.8 | Improved |
| train part and object segmentation separately instead of jointly | LGFormer · PartImageNet val | part mIoU: 67.4 -> 69.4 | Improved |
| evaluate on 20-40% occluded images | LGFormer · Occluded-PartImageNet-v1 | part/object mIoU: part -8.0, object -15.5 | Hurt |
| LGFormer vs Compositor under joint training | Pascal-Part · Pascal-Part val | object mIoU: 83.1 -> 85.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Compositor — Compositor: Bottom-Up Clustering and Compositing for Robust Part and Object Segmentation (CVPR 2023) · 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
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.89)
- Self-Supervised Learning of Object Parts for Semantic Segmentation (CVPR 2022 · similarity 0.89)
- Towards Open-World Segmentation of Parts (CVPR 2023 · similarity 0.89)
- unMORE: Unsupervised Multi-Object Segmentation via Center-Boundary Reasoning (ICML 2025 · similarity 0.87)
- Fine-Grained Image-Text Correspondence with Cost Aggregation for Open-Vocabulary Part Segmentation (CVPR 2025 · similarity 0.87)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.87)
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.
