Research map / Detection and segmentation / Open-set and few-shot detection
Unsupervised learning of object frames by dense equivariant image labelling
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 |
|---|---|---|---|
| use the robust distance loss with γ = 0.5 instead of Llog (ablation) | SIMPLE dense labelling network · MAFL | nearest landmark error: 75.02 % -> 13.29 % | Improved |
| use DILATIONS instead of SIMPLE (ablation) | Ldist, γ = 0.5 dense labelling network · MAFL | nearest landmark error: 13.29 % -> 10.53 % | Improved |
| use DILATIONS instead of SIMPLE (ablation) | Ldist, γ = 0.5 dense labelling network · MAFL | regression landmark error: 7.18 % -> 5.87 % | Improved |
| use DILATIONS instead of SIMPLE (ablation) | Llog dense labelling network · MAFL | nearest landmark error: 75.02 % -> 11.05 % | Improved |
| use DILATIONS instead of SIMPLE (ablation) | Llog dense labelling network · MAFL | regression landmark error: — -> 5.83 % | Improved |
| use the distance loss with γ = 0.5 instead of the negative log-likelihood loss (ablation) | DILATIONS dense labelling network · MAFL | regression landmark error: 5.83 % -> 5.87 % | No clear effect |
| use the distance loss with γ = 0.5 instead of Llog (ablation) | SIMPLE dense labelling network · Roboarm | capsule center prediction error (% image width): 0.97 % -> 1.14 % | Hurt |
| use label dimension 2 instead of 3 (ablation) | SIMPLE Ldist, γ = 0.5 · Roboarm | capsule center prediction error (% image width): 1.14 % -> 1.29 % | Hurt |
| use label dimension 5 instead of 3 (ablation) | SIMPLE Ldist, γ = 0.5 · Roboarm | capsule center prediction error (% image width): 1.14 % -> 1.16 % | No clear effect |
| use label dimension 20 instead of 3 (ablation) | SIMPLE Ldist, γ = 0.5 · Roboarm | capsule center prediction error (% image width): 1.14 % -> 1.28 % | Hurt |
| increase the receptive field by switching from SIMPLE to DILATIONS (ablation) | dense face labelling · Human faces | left/right eye disambiguation | Improved |
| use Llog rather than the distance loss (ablation) | SIMPLE dense labelling network · Human faces | landmark consistency | Hurt |
Closest papers by meaning
- Continuous Surface Embeddings (NeurIPS 2020 · similarity 0.88)
- Equivariant Neural Rendering (ICML 2020 · similarity 0.87)
- Self-Supervised Category-Level Articulated Object Pose Estimation with Part-Level SE(3) Equivariance (ICLR 2023 · similarity 0.87)
- Canonical 3D Deformer Maps: Unifying parametric and non-parametric methods for dense weakly-supervised category reconstruction (NeurIPS 2020 · similarity 0.87)
- Unsupervised Learning of Category-Level 3D Pose from Object-Centric Videos (CVPR 2024 · similarity 0.86)
- Neural Isometries: Taming Transformations for Equivariant ML (NeurIPS 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.
