Hyperbolic Image Embeddings

@article{Khrulkov2020HyperbolicIE,
  title={Hyperbolic Image Embeddings},
  author={Valentin Khrulkov and Leyla Mirvakhabova and E. Ustinova and I. Oseledets and Victor S. Lempitsky},
  journal={2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2020},
  pages={6417-6427}
}
Computer vision tasks such as image classification, image retrieval, and few-shot learning are currently dominated by Euclidean and spherical embeddings so that the final decisions about class belongings or the degree of similarity are made using linear hyperplanes, Euclidean distances, or spherical geodesic distances (cosine similarity). In this work, we demonstrate that in many practical scenarios, hyperbolic embeddings provide a better alternative. 

Figures and Tables from this paper

Hyperbolic Image Segmentation
TLDR
This work shows that hyperbolic manifolds provide a valuable alternative for image segmentation and proposes a tractable formulation of hierarchical pixel-level classification inHyperbolic Image Segmentation, which opens up new possibilities and practical benevolence for segmentation.
On Hyperbolic Embeddings in 2D Object Detection
TLDR
This work incorporates a hyperbolic classifier in two-stage, keypoint-based, and transformerbased object detection architectures and evaluates them on large-scale, long-tailed, and zero-shot object detection benchmarks, resulting in lower classification errors and boosting the overall object detection performance.
Unsupervised Hyperbolic Metric Learning
TLDR
This work proposes an Unsupervised Hyperbolic Metric Learning method with Hierarchical Similarity that considers the natural hierarchies of data by taking advantage ofHyperbolic metric learning and hierarchical clustering, which can effectively excavate richer similarity information beyond binary in modeling.
Hyperbolic Vision Transformers: Combining Improvements in Metric Learning
TLDR
At the core of the method is a vision transformer with output embeddings mapped to hyperbolic space that is directly optimized using modified pairwise cross-entropy loss.
Adaptive Poincaré Point to Set Distance for Few-Shot Classification
TLDR
This paper proposes to learn a context-aware hyperbolic metric to characterize the distance between a point and a set associated with a learned set to set distance and empirically shows that such metric yields robustness in the presence of outliers and achieves a tangible improvement over baseline models.
Adaptive Poincar\'e Point to Set Distance for Few-Shot Classification
TLDR
This paper proposes to learn a context-aware hyperbolic metric to characterize the distance between a point and a set associated with a learned set to set distance and empirically shows that such metric yields robustness in the presence of outliers and achieves a tangible improvement over baseline models.
Non-linear Embeddings in Hilbert Simplex Geometry
TLDR
The findings demonstrate that Hilbert simplex geometry is competitive to alternative geometries such as the Poincaré hyperbolic ball or the Euclidean geometry for embedding tasks while being fast and numerically robust.
Free Hyperbolic Neural Networks with Limited Radii
TLDR
An empirical study is conducted showing that the inferior performance of HNNs on standard recognition datasets can be attributed to the notorious vanishing gradient problem, and a simple yet effective solution called Feature Clipping, which regularizes the hyperbolic embedding whenever its norm exceeding a given threshold is reached.
Connecting Sphere Manifolds Hierarchically for Regularization
TLDR
This technique replaces the last layer of a neural network by combining a spherical fully-connected layer with a hierarchical layer, which is shown to improve the performance of widely used deep neural network architectures on publicly available datasets.
Clipped Hyperbolic Classifiers Are Super-Hyperbolic Classifiers
TLDR
This work proposes an effective solution to HNNs’ poorer general classification performance results from vanishing gradients during backpropagation by simply clipping the Euclidean feature magnitude while training Hnns, and demonstrates that clipped HNNS become super-hyperbolic classifiers.
...
...

References

SHOWING 1-10 OF 76 REFERENCES
ImageNet Large Scale Visual Recognition Challenge
TLDR
The creation of this benchmark dataset and the advances in object recognition that have been possible as a result are described, and the state-of-the-art computer vision accuracy with human accuracy is compared.
Poincaré Embeddings for Learning Hierarchical Representations
TLDR
This work introduces a new approach for learning hierarchical representations of symbolic data by embedding them into hyperbolic space -- or more precisely into an n-dimensional Poincare ball -- and introduces an efficient algorithm to learn the embeddings based on Riemannian optimization.
FaceNet: A unified embedding for face recognition and clustering
TLDR
A system that directly learns a mapping from face images to a compact Euclidean space where distances directly correspond to a measure offace similarity, and achieves state-of-the-art face recognition performance using only 128-bytes perface.
Discriminative k-shot learning using probabilistic models
TLDR
It is shown that even a simple probabilistic model achieves state-of-the-art on a standard k-shot learning dataset by a large margin and is able to accurately model uncertainty, leading to well calibrated classifiers, and is easily extensible and flexible, unlike many recent approaches to k- shot learning.
Sampling Matters in Deep Embedding Learning
TLDR
This paper proposes distance weighted sampling, which selects more informative and stable examples than traditional approaches, and shows that a simple margin based loss is sufficient to outperform all other loss functions.
Revisiting Local Descriptor Based Image-To-Class Measure for Few-Shot Learning
TLDR
This work proposes a Deep Nearest Neighbor Neural Network (DN4), a simple, effective, and computationally efficient framework for few-shot learning that not only learns the optimal deep local descriptors for the image-to-class measure, but also utilizes the higher efficiency of such a measure in the case of example scarcity.
Few-Shot Image Recognition by Predicting Parameters from Activations
TLDR
A novel method that can adapt a pre-trained neural network to novel categories by directly predicting the parameters from the activations is proposed, which achieves the state-of-the-art classification accuracy on Novel categories by a significant margin while keeping comparable performance on the large-scale categories.
Spot and Learn: A Maximum-Entropy Patch Sampler for Few-Shot Image Classification
TLDR
This work proposes a sampling method that de-correlates an image based on maximum entropy reinforcement learning, and extracts varying sequences of patches on every forward-pass with discriminative information observed, resulting in improved FSL performances.
Very Deep Convolutional Networks for Large-Scale Image Recognition
TLDR
This work investigates the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting using an architecture with very small convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers.
Deep Metric Learning via Lifted Structured Feature Embedding
TLDR
An algorithm for taking full advantage of the training batches in the neural network training by lifting the vector of pairwise distances within the batch to the matrix of Pairwise distances enables the algorithm to learn the state of the art feature embedding by optimizing a novel structured prediction objective on the lifted problem.
...
...