Research map / Detection and segmentation / Open-set and few-shot detection
Unsupervised Learning of Object Landmarks by Factorized Spatial Embeddings
James Thewlis, Hakan Bilen, Andrea Vedaldi
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 |
|---|---|---|---|
| increase number of unsupervised landmarks from 10 to 50 (ablation) | Unsupervised landmarks · MAFL test (regressor on MAFL) | mean error: 7.95 -> 6.67 | Improved |
| increase number of unsupervised landmarks from 10 to 50 (ablation) | Unsupervised landmarks · MAFL test (regressor on CelebA) | mean error: 6.32 -> 5.33 | Improved |
| train the regressor on CelebA instead of MAFL (ablation) | Unsupervised landmarks, 50 landmarks · MAFL test | mean error: 6.67 -> 5.33 | Improved |
| use all 19,000 MAFL training images instead of 10 (ablation) | Unsupervised landmarks, 30 landmarks · MAFL test | mean error: 8.49 -> 7.15 | Improved |
| use 20 labelled images instead of 1 (ablation) | Unsupervised landmarks, 30 landmarks · MAFL test | mean error: 10.82 -> 8.06 | Improved |
| add unsupervised pretraining before regressing on 5 AFLW images (ablation) | Landmark regressor · AFLW (5 pts) | mean error: 23.85 -> 12.94 | Improved |
| add unsupervised fine-tuning on target 300-W data (ablation) | Unsupervised landmarks, 50 landmarks · 300-W (68 pts) | mean error: 9.30 -> 7.97 | Improved |
| replace fully supervised detectors with unsupervised landmarks plus linear regressor | Unsupervised landmarks, 50 points · MAFL test | mean error: 7.95 -> 6.67 | Improved |
| replace fully supervised detectors with unsupervised landmarks plus linear regressor | Unsupervised landmarks, 51 points · AFLW (5 pts) | mean error: 7.23 -> 10.53 | Hurt |
| replace fully supervised RAR with unsupervised landmarks plus linear regressor | Unsupervised landmarks, 50 landmarks · 300-W (68 pts) | mean error: 4.94 -> 7.97 | Hurt |
Closest papers by meaning
- Object landmark discovery through unsupervised adaptation (NeurIPS 2019 · similarity 0.93)
- Unsupervised Learning of Object Landmarks via Self-Training Correspondence (NeurIPS 2020 · similarity 0.93)
- Unsupervised Discovery of Object Landmarks as Structural Representations (CVPR 2018 · similarity 0.92)
- Unsupervised Learning of Object Landmarks through Conditional Image Generation (NeurIPS 2018 · similarity 0.92)
- From Keypoints to Object Landmarks via Self-Training Correspondence: A novel approach to Unsupervised Landmark Discovery (TPAMI 2023 · similarity 0.90)
- Unsupervised 3D Structure Inference from Category-Specific Image Collections (CVPR 2024 · 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.
