Research map / Detection and segmentation / Open-set and few-shot detection
PartDistillation: Learning Parts From Instance Segmentation
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 |
|---|---|---|---|
| Use the two-stage PartDistillation pipeline instead of one-stage self-training | Part segmentation from ImageNet-21K · Pascal Part (zero-shot, no GT object masks) | AR@200: 18.8 -> 25.0 | Helped |
| Use the two-stage PartDistillation pipeline instead of one-stage self-training | Part segmentation from ImageNet-21K · Pascal Part (zero-shot, with GT object masks) | mIoU: 16.5 -> 23.0 | Helped |
| Use PartDistillation rather than supervised transfer from Pascal Part | Zero-shot part segmentation · PartImageNet (with GT object masks) | AR@200: 45.7 -> 58.0 | Helped |
| Initialize with PartDistillation rather than COCO instance segmentation | Few-shot part segmentation · Pascal Part (1% labels) | AR@200: 25.3 -> 33.2 | Helped |
| Initialize with PartDistillation rather than COCO instance segmentation | Few-shot part segmentation · PartImageNet (1% labels) | mIoU: 32.0 -> 36.3 | Helped |
| Initialize with PartDistillation rather than COCO instance segmentation | Few-shot part segmentation · Cityscapes Part (1% labels) | mIoU: 28.3 -> 53.4 | Helped |
| Use COCO Mask2Former features rather than ImageNet-21K classification features for pixel grouping (ablation) | SwinL pixel grouping · PartImageNet-test | AR@10: 9.4 -> 27.6 | Helped |
| Use multi-scale features (ablation) | Mask2Former pixel grouping · PartImageNet-test | AR@10: 9.8 -> 16.1 | Helped |
| Use instance-level clustering (ablation) | Mask2Former pixel grouping · PartImageNet-test | AR@10: 16.1 -> 23.4 | Helped |
| Use cosine similarity for clustering (ablation) | Mask2Former pixel grouping · PartImageNet-test | AR@10: 14.4; 21.5; 23.4; 27.6 | Mixed |
| Combine multi-scale features, cosine distance, and instance-level clustering (ablation) | Mask2Former pixel grouping · PartImageNet-test | AR@10: 23.4 -> 27.6 | Helped |
Closest papers by meaning
- PartDistill: 3D Shape Part Segmentation by Vision-Language Model Distillation (CVPR 2024 · similarity 0.88)
- Fine-Grained Image-Text Correspondence with Cost Aggregation for Open-Vocabulary Part Segmentation (CVPR 2025 · similarity 0.88)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.87)
- PartNeXt: A Next-Generation Dataset for Fine-Grained and Hierarchical 3D Part Understanding (NeurIPS 2025 · similarity 0.87)
- Understanding Multi-Granularity for Open-Vocabulary Part Segmentation (NeurIPS 2024 · similarity 0.87)
- PARTONOMY: Large Multimodal Models with Part-Level Visual Understanding (NeurIPS 2025 · 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.