Research map / Detection and segmentation / CNN features
PDiscoNet: Semantically consistent part discovery for fine-grained recognition
Robert van der Klis, Stephan Alaniz, Massimiliano Mancini, Cassio F. Dantas, Dino Ienco, Zeynep Akata, Diego Marcos
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 |
|---|---|---|---|
| add orthogonality loss (ablation) | PDiscoNet · CUB (K=16) | NMI: 36.12 -> 56.87 | Improved |
| add equivariance loss (ablation) | PDiscoNet · CUB (K=16) | NMI: 40.22 -> 56.87 | Improved |
| add part dropout (ablation) | PDiscoNet · CUB (K=16) | NMI: 46.37 -> 56.87 | Improved |
| add part vector modulation (ablation) | PDiscoNet · CUB (K=16) | NMI: 53.45 -> 56.87 | Improved |
| add concentration loss (ablation) | PDiscoNet · CUB (K=16) | Keypoint regression error: 8.58 -> 7.67 | Improved |
| add presence loss on CUB (ablation) | PDiscoNet · CUB (K=16) | NMI: 55.18 -> 56.87 | No clear effect |
| add presence loss on PartImageNet (ablation) | PDiscoNet · PartImageNet (K=8) | NMI: 12.22 -> 27.13 | Improved |
| add auxiliary losses on PartImageNet (ablation) | PDiscoNet · PartImageNet (K=8) | classification accuracy: 90.32 (no conc) vs 88.58 (full) | Hurt |
| increase part dropout rate from 0.3 to 0.9 (ablation) | PDiscoNet · CUB (K=8) | NMI / classification accuracy: NMI 49.22 -> 54.42; accuracy 87.31 -> 83.34 | Mixed |
| increase number of parts K from 4 to 16 (ablation) | PDiscoNet · CUB | classification accuracy: 86.17 -> 87.49 | Improved |
| use PDiscoNet instead of Huang and Li under Gaussian noise | CUB (K=16) · CUB with Gaussian noise | classification accuracy: 22.1 -> 38.2 at noise SD 1.50 | Improved |
Later papers that compared with this method
1 later accepted paper reports PDiscoNet as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- Unsupervised Part Discovery via Descriptor-Based Masked Image Restoration with Optimized Constraints (ICCV 2025 · better on 0 of 1 shared result)
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
- PDiscoFormer: Relaxing Part Discovery Constraints with Vision Transformers (ECCV 2024 · similarity 0.90)
- Towards Open-World Segmentation of Parts (CVPR 2023 · similarity 0.88)
- Adaptive Part Learning for Fine-Grained Generalized Category Discovery: A Plug-and-Play Enhancement (CVPR 2025 · similarity 0.87)
- Multi-Part Object Representations via Graph Structures and Co-Part Discovery (NeurIPS 2026 · similarity 0.87)
- Fine-Grained Image-Text Correspondence with Cost Aggregation for Open-Vocabulary Part Segmentation (CVPR 2025 · similarity 0.86)
- Interpretable Image Classification via Non-parametric Part Prototype Learning (CVPR 2025 · 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.
