CVPR Findings2026
Equivariant Unsupervised Object Detection with Learnable Riesz Transform and Composite Spatial Transformers
Sayan Kumar Chaki, Thierry Fournel, Rémi Emonet
Abstract
Building models robust to transformations such as ro tation, scale, and translation is a challenge in machine learning and computer vision. Existing approaches often provide only partial and discrete equivariance (group equivariance) or rely on supervision or very abundant data to learn equivariant representations. To achieve finegrained equivariance from low data, we combine and improve over both approaches. We propose a novel, learnable, Riesztransformbased ar chitecture that achieves builtin group equivariance for translation, rotation, and scale. We combine it with a Spatial Transform Network (STN) tailored for the sequential estimation of composite transformations, reducing the combinatorial data requirements for learn ing finegrained equivariance. Improved generalization guarantees and extensive experiments demonstrate that our approach brings improvements over stateoftheart methods in unsupervised representation learning and object discovery, even more so in lowdata regimes.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
Builds on12
- SPACE: Unsupervised Object-Oriented Scene Representation via Spatial Attention and DecompositionZhixuan Lin, Yi-Fu Wu, Skand Vishwanath Peri, Weihao Sun et al.ICLR 2020 · 276 citations
- Reducing SO(3) Convolutions to SO(2) for Efficient Equivariant GNNsSaro Passaro, C. Lawrence ZitnickICML 2023 · 157 citations
- Deep Clustering by Gaussian Mixture Variational Autoencoders With Graph EmbeddingLinxiao Yang, Ngai-Man Cheung, Jiaying Li, Jun FangICCV 2019 · 149 citations
- A Program to Build E(N)-Equivariant Steerable CNNsGabriele Cesa, Leon Lang, Maurice WeilerICLR 2022 · 133 citations
- Spherical Channels for Modeling Atomic InteractionsLarry Zitnick, Abhishek Das, Adeesh Kolluru, Janice Lan et al.NeurIPS 2022 · 84 citations
Related papers
- Unsupervised Learning of Group Invariant and Equivariant RepresentationsRobin Winter, Marco Bertolini, Tuan Le, Frank Noé et al.NeurIPS 2022 · 61 citations
- Empowering Networks With Scale and Rotation Equivariance Using A Similarity ConvolutionZikai Sun, Thierry BluICLR 2023 · 2 citations
- Scale-Equivariant Steerable NetworksIvan Sosnovik, Michal Szmaja, Arnold W. M. SmeuldersICLR 2020 · 169 citations
- Unsupervised Object Representation Learning using Translation and Rotation Group Equivariant VAEAlireza Nasiri, Tristan BeplerNeurIPS 2022 · 18 citations
- Learning SO(3)-Invariant Semantic Correspondence via Local Shape TransformChunghyun Park, Seungwook Kim, Jaesik Park, Minsu ChoCVPR 2024 · 2 citations
