Research map / Detection and segmentation / Open-set and few-shot detection
Unsupervised Discovery of Object Landmarks as Structural Representations
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 differentiable landmark discovery with reconstruction and validity constraints | Unsupervised landmark discovery · MAFL test | mean landmark error: 6.32 -> 3.46 | Helped |
| Increase discovered landmarks from 10 to 30 | Ours · MAFL test | mean landmark error: 3.46 -> 3.15 | Helped |
| Add reconstruction loss (ablation) | Full landmark discovery method · MAFL test | mean landmark error: 3.45 -> 3.15 | Helped |
| Add concentration loss (ablation) | Full landmark discovery method · MAFL test | mean landmark error: 3.91 -> 3.15 | Helped |
| Add separation loss (ablation) | Full landmark discovery method · MAFL test | mean landmark error: 16.56 -> 3.15 | Helped |
| Add equivariance loss (ablation) | Full landmark discovery method · MAFL test | mean landmark error: 8.42 -> 3.15 | Helped |
| Use optical-flow self-supervision for equivariance (ablation) | Ours · Human3.6M | mean landmark prediction error: 4.91 -> 4.14 | Helped |
| Include discovered landmarks as TPS control points alongside grid points (ablation) | Ours · MAFL test | mean landmark prediction error: 4.17 -> 3.46 | Helped |
| Vary the grid-to-landmark TPS control-point ratio (ablation) | Ours · MAFL test | mean landmark prediction error: 3.38 to 3.86 | No clear effect |
| Use discovered landmarks for annotated-landmark prediction instead of the previous unsupervised method | 30-landmark discovery · MAFL test, CelebA regressor training | mean landmark error: 5.76 -> 3.15 | Helped |
| Use discovered landmarks for annotated-landmark prediction instead of the previous unsupervised method | 24-landmark discovery · Car test | mean landmark error: 11.11 -> 5.80 | Helped |
| Use discovered landmarks for annotated-landmark prediction instead of the previous unsupervised method | 20-landmark discovery · Cat head test | mean landmark error: 26.94 -> 14.84 | Helped |
Closest papers by meaning
- Unsupervised Learning of Object Landmarks via Self-Training Correspondence (NeurIPS 2020 · similarity 0.90)
- Object landmark discovery through unsupervised adaptation (NeurIPS 2019 · similarity 0.90)
- Unsupervised Learning of Object Landmarks through Conditional Image Generation (NeurIPS 2018 · similarity 0.89)
- Self-Supervised Learning of Geometrically Stable Features Through Probabilistic Introspection (CVPR 2018 · similarity 0.88)
- Unsupervised Learning of 3D Structure from Images (NeurIPS 2016 · similarity 0.86)
- Pose-Guided Self-Training with Two-Stage Clustering for Unsupervised Landmark Discovery (CVPR 2024 · similarity 0.85)
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.