Research map / Detection and segmentation / CNN features
Unsupervised Part Discovery via Descriptor-Based Masked Image Restoration with Optimized Constraints
Jiahao Xia, Yike Wu, Wenjian Huang, Jianguo Zhang, Jian Zhang
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 foreground loss L_f (ablation) | MPAE · PartImageNet Segmentation (K=50) | NMI (%): 0.00 -> 55.10 | Improved |
| add background loss L_b (ablation) | MPAE · PartImageNet Segmentation (K=50) | NMI (%): 39.36 -> 55.10 | Improved |
| add total variation loss L_v (ablation) | MPAE · PartImageNet Segmentation (K=50) | NMI (%): 14.37 -> 55.10 | Improved |
| add entropy loss L_e (ablation) | MPAE · PartImageNet Segmentation (K=50) | NMI (%): 52.51 -> 55.10 | Improved |
| add reconstruction loss L_r (ablation) | MPAE · PartImageNet Segmentation (K=50) | NMI (%): 22.32 -> 55.10 | Improved |
| add semantic constraint L_s (ablation) | MPAE · PartImageNet Segmentation (K=50) | NMI (%): 56.42 -> 55.10 | Mixed |
| add semantic constraint L_s (ablation) | MPAE · PartImageNet Segmentation (K=50) | average discovered parts per image: 3.94 -> 9.27 | Improved |
| raise masking ratio from 50% to 90% (ablation) | MPAE · PartImageNet Segmentation (K=50) | NMI (%): 45.50 -> 55.10 | Improved |
| use masking ratio 100% instead of 90% (ablation) | MPAE · PartImageNet Segmentation (K=50) | NMI (%): 55.10 -> 43.39 | Hurt |
| add structural difference penalty (VGG-19) to L_r (ablation) | MPAE · PartImageNet Segmentation (K=50) | NMI (%): 19.65 -> 55.10 | Improved |
| increase encoder layers from 1 to 2 (ablation) | MPAE · PartImageNet Segmentation (K=50) | NMI (%): 50.71 -> 55.10 | Improved |
| increase decoder layers from 2 to 6 (ablation) | MPAE · PartImageNet Segmentation (K=50) | NMI (%): 55.10 -> 52.69 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- PDiscoNet — PDiscoNet: Semantically consistent part discovery for fine-grained recognition (ICCV 2023) · 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
- Unsupervised Part Discovery via Dual Representation Alignment (TPAMI 2024 · similarity 0.91)
- Unsupervised Part Discovery from Contrastive Reconstruction (NeurIPS 2021 · similarity 0.90)
- Unsupervised Part Segmentation Through Disentangling Appearance and Shape (CVPR 2021 · similarity 0.89)
- Unsupervised Part-Based Disentangling of Object Shape and Appearance (CVPR 2019 · similarity 0.89)
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 2025 · similarity 0.86)
- Multi-Part Object Representations via Graph Structures and Co-Part Discovery (NeurIPS 2026 · 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.
