Geometric and Representation Learning Advances

The field is witnessing a significant shift towards leveraging geometric and representation learning techniques to tackle complex problems. Researchers are exploring innovative methods to improve image retargeting, surface parameterization, and optimal transport in various domains, including computer vision and graphics. Notably, the use of self-supervised learning, mesh deformation, and quasiconformal maps is becoming increasingly popular. Additionally, there is a growing interest in developing efficient algorithms for high-dimensional diffeomorphic mapping, tensor decomposition, and MRI representation learning. These advances have the potential to revolutionize various applications, including image processing, computer-aided design, and medical imaging. Noteworthy papers include: Object-IR, which achieves state-of-the-art performance in image retargeting by leveraging object consistency and mesh deformation. Hyperbolic Optimal Transport, which proposes a novel algorithm for computing optimal transport maps in hyperbolic space. Metadata-Aligned 3D MRI Representations, which introduces a metadata-guided framework for learning MRI contrast representations. No-rank Tensor Decomposition Using Metric Learning, which presents a framework for tensor decomposition grounded in metric learning. Distribution-Aware Tensor Decomposition for Compression of Convolutional Neural Networks, which proposes a data-informed approach for compressing neural networks.

Sources

Object-IR: Leveraging Object Consistency and Mesh Deformation for Self-Supervised Image Retargeting

Surface parameterization via optimization of relative entropy and quasiconformality

Hyperbolic Optimal Transport

Metadata-Aligned 3D MRI Representations for Contrast Understanding and Quality Control

No-rank Tensor Decomposition Using Metric Learning

Variational Geometry-aware Neural Network based Method for Solving High-dimensional Diffeomorphic Mapping Problems

Wavelet-Optimized Motion Artifact Correction in 3D MRI Using Pre-trained 2D Score Priors

Vectorized Computation of Euler Characteristic Functions and Transforms

Normalized tensor train decomposition

Distribution-Aware Tensor Decomposition for Compression of Convolutional Neural Networks

Built with on top of