Research map / Image generation / Autoregressive and latent generation
Unsupervised Learning of Object Landmarks through Conditional Image Generation
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 spatial keypoint bottleneck rather than a fully connected bottleneck (ablation) | Conditional image generation landmark detector · MAFL | normalized % MSE: 20.60 -> 2.58 | Helped |
| Use the content/perceptual reconstruction loss rather than pixel L1 loss (ablation) | Conditional image generation landmark detector · MAFL | normalized % MSE: 3.64 -> 2.58 | Helped |
| Use the content/perceptual reconstruction loss rather than adversarial plus pixel L1 loss (ablation) | Conditional image generation landmark detector · MAFL | normalized % MSE: 3.62 -> 2.58 | Helped |
| Use the content/perceptual reconstruction loss rather than pixel L2 loss (ablation) | Conditional image generation landmark detector · MAFL | normalized % MSE: 2.84 -> 2.58 | Helped |
| Use the content/perceptual reconstruction loss rather than adversarial plus pixel L2 loss (ablation) | Conditional image generation landmark detector · MAFL | normalized % MSE: 2.80 -> 2.58 | Helped |
| Add a bias term to the linear regressor | VoxCeleb-trained detector on MAFL · MAFL | normalized % MSE: 3.94 -> 3.63 | Helped |
| Train on VoxCeleb video frames rather than CelebA images | Landmark detector · MAFL | normalized % MSE: 2.58 -> 3.94 | Hurt |
| Use a supervised rather than self-supervised perceptual-loss network | CelebA-trained detector, K=30 · MAFL | normalized % MSE: 2.58 -> 2.63 | Hurt |
| Use a supervised rather than self-supervised perceptual-loss network | CelebA-trained detector, K=30 · AFLW | normalized % MSE: 6.31 -> 6.39 | Hurt |
| Use 30 unsupervised landmarks rather than 10 | CelebA-trained detector · MAFL | normalized % MSE: 3.19 -> 2.58 | Helped |
Closest papers by meaning
- Unsupervised Learning of Object Landmarks via Self-Training Correspondence (NeurIPS 2020 · similarity 0.91)
- Unsupervised Discovery of Object Landmarks as Structural Representations (CVPR 2018 · similarity 0.89)
- Object landmark discovery through unsupervised adaptation (NeurIPS 2019 · similarity 0.89)
- Self-Supervised Learning of Geometrically Stable Features Through Probabilistic Introspection (CVPR 2018 · similarity 0.87)
- Unsupervised Discovery of Facial Landmarks and Head Pose (CVPR 2025 · similarity 0.87)
- Style Aggregated Network for Facial Landmark Detection (CVPR 2018 · 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.