Research map / Detection and segmentation / Open-set and few-shot detection
Unsupervised Object Representation Learning using Translation and Rotation Group Equivariant VAE
Alireza Nasiri, Tristan Bepler
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 discrete rotations from P4 to P16 | TARGET-VAE · MNIST(U) | rotation correlation: 0.80 -> 0.93 | Improved |
| increase discrete rotations from P4 to P16 | TARGET-VAE · MNIST(U) | translation correlation: 0.975, 0.976 -> 0.974, 0.971 | No clear effect |
| use TARGET-VAE instead of Spatial-VAE | TARGET-VAE P8 · MNIST(U) | rotation correlation: 0.005 -> 0.859 | Improved |
| use TARGET-VAE instead of Spatial-VAE | TARGET-VAE P8 · MNIST(N) | translation correlation: 0.994, 0.993 -> 0.981, 0.981 | Hurt |
| increase z_dim from 2 to 32 | TARGET-VAE P8 · MNIST(N) | clustering accuracy (%): 60.1 -> 77.7 | Improved |
| increase z_dim from 2 to 32 | TARGET-VAE P16 · MNIST(U) | clustering accuracy (%): 63.4 -> 71.2 | Improved |
| increase discrete rotations from P8 to P16 (z_dim=32) | TARGET-VAE · MNIST(N) | clustering accuracy (%): 77.7 -> 75.2 | Hurt |
| replace Spatial-VAE with TARGET-VAE P4 (z_dim=2) | TARGET-VAE · MNIST(U) | clustering accuracy (%): 28.2 -> 56.6 | Improved |
| replace group conv with regular conv, inference only on translation (Variant 1) (ablation) | TARGET-VAE P4 · MNIST(U) | clustering accuracy (%): 56.6 -> 33.8 | Hurt |
| use group conv layers without rotation inference (Variant 2) (ablation) | Variant 1 · MNIST(U) | clustering accuracy (%): 33.8% -> 37.1% | Improved |
| remove theta_offset from posterior on theta (Variant 3) (ablation) | TARGET-VAE P4 · MNIST(U) | rotation correlation: 0.80 -> 0.008 | Hurt |
| remove theta_offset from posterior on theta (Variant 3) (ablation) | TARGET-VAE P4 · MNIST(U) | clustering accuracy (%): 56.6% -> 36.6% | Hurt |
Closest papers by meaning
- Explicitly disentangling image content from translation and rotation with spatial-VAE (NeurIPS 2019 · similarity 0.90)
- Unsupervised Representation Learning by Predicting Image Rotations (ICLR 2018 · similarity 0.88)
- Shape-Pose Disentanglement Using SE(3)-Equivariant Vector Neurons (ECCV 2022 · similarity 0.87)
- Rotation and Translation Invariant Representation Learning with Implicit Neural Representations (ICML 2023 · similarity 0.87)
- R2Det: Exploring Relaxed Rotation Equivariance in 2D Object Detection (ICLR 2025 · similarity 0.85)
- One-shot 3D Object Canonicalization based on Geometric and Semantic Consistency (CVPR 2025 · 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.
