Research map / Label-efficient and robust learning / Recognition and long-tail
Deep Transformation-Invariant Clustering
Tom Monnier, Thibault Groueix, Mathieu Aubry
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 |
|---|---|---|---|
| Add transformation-invariant alignment with DTI K-means rather than pixel-space K-means | K-means · MNIST | ACC: 54.8 -> 97.3 | Improved |
| Add transformation-invariant alignment with DTI K-means rather than pixel-space K-means | K-means · Fashion-MNIST | ACC: 54.1 -> 61.2 | Improved |
| Use Gaussian-weighted clustering loss focused on the center digit | DTI clustering on SVHN · SVHN | ACC: 36.4 / 44.5 -> 39.6 / 62.6 | Improved |
| Add morphological transformation (ablation) | DTI clustering with affine and TPS · MNIST | ACC: 94.8 -> 97.3 | Improved |
| Add thin plate spline transformation (ablation) | DTI clustering with affine and morphological transformation · MNIST | ACC: 90.0 -> 97.3 | Improved |
| Add affine transformation (ablation) | DTI clustering with morphological and TPS transformations · MNIST | ACC: 85.1 -> 97.3 | Improved |
| Order modules affine, TPS, then morphological instead of affine, morphological, then TPS (ablation) | DTI clustering · MNIST | ACC: 97.3 -> 95.5 | Hurt |
| Order modules morphological, affine, then TPS instead of affine, morphological, then TPS (ablation) | DTI clustering · MNIST | ACC: 97.3 -> 27.5 | Hurt |
| Add empty cluster reassignment (ablation) | DTI clustering · MNIST | ACC: 80.9 -> 97.3 | Improved |
| Add curriculum learning for transformation modules (ablation) | DTI clustering · MNIST | ACC: 83.9 -> 97.3 | Improved |
| Use learned transformation predictors instead of directly optimizing per-sample transformations (ablation) | Transformation-invariant clustering · MNIST | ACC: 26.3 -> 97.3 | Improved |
| Use learned transformation predictors instead of directly optimizing per-sample affine transformations (ablation) | Transformation-invariant clustering · MNIST | ACC: 73.0 -> 97.3 | Improved |
Closest papers by meaning
- PointClustering: Unsupervised Point Cloud Pre-Training Using Transformation Invariance in Clustering (CVPR 2023 · similarity 0.88)
- Deep Clustering for Unsupervised Learning of Visual Features (ECCV 2018 · similarity 0.88)
- Invariant Information Clustering for Unsupervised Image Classification and Segmentation (ICCV 2019 · similarity 0.88)
- Deep Clustering via Joint Convolutional Autoencoder Embedding and Relative Entropy Minimization (ICCV 2017 · similarity 0.87)
- ManlyMM-VAE: Deep Clustering With Manly Transformation Mixture Latent Embeddings for High-Dimensional Skewed Data (TPAMI 2026 · similarity 0.87)
- Image Clustering via the Principle of Rate Reduction in the Age of Pretrained Models (ICLR 2024 · similarity 0.86)
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.
