Research map / Label-efficient and robust learning / Recognition and long-tail
Rotation and Translation Invariant Representation Learning with Implicit Neural Representations
Sehyun Kwon, Joo Young Choi, Ernest K. Ryu
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 IRL-INR semantic representations as the SCAN pretext task instead of SimCLR | SCAN · MNIST(U) | clustering accuracy: 85.4 ± 1.46 -> 90.4 ± 1.74 | Improved |
| Use IRL-INR semantic representations as the SCAN pretext task instead of SimCLR | SCAN · WM811k | clustering accuracy: 57.1 ± 2.81 -> 64.6 ± 1.01 | Improved |
| Increase semantic dimension from d=32 to d=512 (ablation) | IRL-INR + SCAN · MNIST(U) | clustering accuracy: 84.18 ± 2.11 -> 90.4 ± 1.74 | Improved |
| Increase semantic dimension from d=32 to d=512 (ablation) | IRL-INR + SCAN · WM811k | clustering accuracy: 53.78 ± 3.41 -> 64.6 ± 1.01 | Improved |
| Increase SimCLR + SCAN backbone output dimension (ablation) | SimCLR + SCAN · MNIST(U) and WM811k | clustering accuracy | No clear effect |
| Increase TARGET-VAE latent dimension (ablation) | TARGET-VAE, with or without SCAN · Clustering benchmarks | clustering accuracy | No clear effect |
| Add the hypernetwork-INR architecture instead of a simple MLP generator or pixel-warped autoencoder (ablation) | IRL-INR · WM811k and WHOI-Plankton | reconstruction quality | Improved |
| Add the consistency loss (ablation) | IRL-INR · Clustering with semantic representations | clustering accuracy | Improved |
| Add the symmetry-breaking loss (ablation) | IRL-INR · Reconstruction and clustering | reconstruction quality | Improved |
| Add random rotation and translation to SimCLR pretext and clustering augmentations (ablation) | SimCLR + SCAN · MNIST(U) | clustering accuracy: 52.8 ± 3.86 -> 85.4 ± 1.46 | Improved |
| Add random translation to rotation augmentation for SimCLR + SCAN (ablation) | SimCLR + SCAN · MNIST(U) | clustering accuracy: 83.66 ± 1.71 -> 85.4 ± 1.46 | Improved |
| Use stronger S1+S2+R+T clustering augmentation instead of R+T (ablation) | IRL-INR + SCAN · MNIST(U) | clustering accuracy: 87.11 ± 1.24 -> 90.4 ± 1.74 | Improved |
Closest papers by meaning
- Unsupervised Object Representation Learning using Translation and Rotation Group Equivariant VAE (NeurIPS 2022 · similarity 0.87)
- Polynomial Implicit Neural Representations for Large Diverse Datasets (CVPR 2023 · similarity 0.87)
- Transformers As Meta-Learners for Implicit Neural Representations (ECCV 2022 · similarity 0.86)
- From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication (ICLR 2024 · similarity 0.85)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.85)
- Inductive Gradient Adjustment for Spectral Bias in Implicit Neural Representations (ICML 2025 · 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.
