Research map / Detection and segmentation / Open-set and few-shot detection
Towards Open-World Segmentation of Parts
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 |
|---|---|---|---|
| Train class-agnostically instead of using part classes (ablation) | Mask2Former part segmentation model · PartImageNet val | AP: 39.71 -> 40.01 | Helped |
| Add an object-mask input channel (ablation) | Class-agnostic Mask2Former · PartImageNet val | AP: 40.01 -> 41.94 | Helped |
| Filter predicted parts outside imperfect object masks (ablation) | Class-agnostic Mask2Former · PartImageNet val | AP: 40.01 -> 42.48 | Helped |
| Combine object-mask input with post-processing (ablation) | Class-agnostic Mask2Former · PartImageNet val | AP: 41.94 -> 45.40 | Helped |
| Use perfect rather than imperfect object masks with pre- and post-awareness (ablation) | Class-agnostic Mask2Former · PartImageNet val | AP: 45.40 -> 87.61 | Helped |
| Add self-supervised pixel-clustering fine-tuning (ablation) | Pre-aware model with imperfect object masks · PartImageNet test | AP: 38.96 -> 40.17 | Helped |
| Add self-training with pseudo-labels (ablation) | Pre-aware model with imperfect object masks · PartImageNet test | AP: 38.96 -> 40.38 | Helped |
| Combine self-supervised fine-tuning and self-training (ablation) | Pre-aware model with imperfect object masks · Pascal-Part-58 val | AP: 20.27 -> 24.02 | Helped |
| Combine self-supervised fine-tuning and self-training (ablation) | Pre-aware model with perfect object masks · Pascal-Part-108 val | AP: 13.40 -> 16.40 | Helped |
| Use two self-training rounds instead of one (ablation) | OPS with imperfect masks · PartImageNet test | AP: 40.43 -> 40.78 | Helped |
| Use two self-training rounds instead of one (ablation) | OPS with perfect masks · PartImageNet test | AP: 83.86 -> 83.86 | No clear effect |
| Use two self-training rounds instead of one (ablation) | OPS with imperfect masks · Pascal-Part-58 | AP: 24.02 -> 23.96 | Hurt |
Closest papers by meaning
- OV-PARTS: Towards Open-Vocabulary Part Segmentation (NeurIPS 2023 · similarity 0.91)
- Fine-Grained Image-Text Correspondence with Cost Aggregation for Open-Vocabulary Part Segmentation (CVPR 2025 · similarity 0.90)
- PARTONOMY: Large Multimodal Models with Part-Level Visual Understanding (NeurIPS 2025 · similarity 0.90)
- Understanding Multi-Granularity for Open-Vocabulary Part Segmentation (NeurIPS 2024 · similarity 0.88)
- Multi-Part Object Representations via Graph Structures and Co-Part Discovery (NeurIPS 2026 · similarity 0.88)
- PartNeXt: A Next-Generation Dataset for Fine-Grained and Hierarchical 3D Part Understanding (NeurIPS 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.