Research map / Label-efficient and robust learning / Recognition and long-tail
Interpretable Transformations With Encoder-Decoder Networks
Daniel E. Worrall, Stephan J. Garbin, Daniyar Turmukhambetov, Gabriel J. Brostow
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 feature-transform-layer invariant features with a CNN instead of a standard CNN with data augmentation | CNN classifier · MNIST-rot | test error (%): 3.50 -> 2.14 | Improved |
| use a CNN encoder instead of an MLP encoder in the proposed model | feature transform layer model · MNIST-rot | test error (%): 4.90 -> 2.14 | Improved |
| use the MLP variant of the proposed model instead of a baseline CNN | baseline CNN · MNIST-rot | test error (%): 5.03 -> 4.90 | Improved |
| add a regularization term that keeps invariant code lengths of transformed encodings equal (ablation) | MNIST-rot classifier · MNIST-rot | test error (%) | Improved |
| add the feature transform layer (block-diagonal rotation matrices) to the network | encoder-decoder · training speed | slowdown: slow down of just 2% | Hurt |
| compare against rotation-specific equivariant networks (RotEqNet, Harmonic Networks, P4CNN) | proposed CNN · MNIST-rot | test error (%): 1.16 vs 2.14 | Hurt |
| classify ModelNet10 from relative-phase invariants of the equivariant code instead of using specialized 3D classifiers | ModelNet10 classifier · ModelNet10 | accuracy (%): 97.14% (VRN Ensemble) vs 90.4% | Hurt |
Closest papers by meaning
- Orthogonality-Enforced Latent Space in Autoencoders: An Approach to Learning Disentangled Representations (ICML 2023 · similarity 0.87)
- Explicit Disentanglement of Appearance and Perspective in Generative Models (NeurIPS 2019 · similarity 0.87)
- Unsupervised Robust Disentangling of Latent Characteristics for Image Synthesis (ICCV 2019 · similarity 0.86)
- Where and What? Examining Interpretable Disentangled Representations (CVPR 2021 · similarity 0.85)
- Feature Space Transfer for Data Augmentation (CVPR 2018 · similarity 0.85)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.84)
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.
