Hyperbolic Image Embeddings

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.

PDF Abstract CVPR 2020 PDF CVPR 2020 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Few-Shot Image Classification CUB 200 5-way 1-shot Hyperbolic ProtoNet Accuracy 60.52 # 32
Few-Shot Image Classification CUB 200 5-way 5-shot Hyperbolic ProtoNet Accuracy 72.22 # 29
Few-Shot Image Classification Mini-Imagenet 5-way (1-shot) Hyperbolic ProtoNet Accuracy 51.57 # 88
Few-Shot Image Classification Mini-Imagenet 5-way (5-shot) Hyperbolic ProtoNet Accuracy 66.27 # 83
Few-Shot Image Classification OMNIGLOT - 1-Shot, 20-way Hyperbolic ProtoNet Accuracy 95.9% # 13
Few-Shot Image Classification OMNIGLOT - 1-Shot, 5-way Hyperbolic ProtoNet Accuracy 99.0 # 8
Few-Shot Image Classification OMNIGLOT - 5-Shot, 20-way Hyperbolic ProtoNet Accuracy 98.15% # 15
Few-Shot Image Classification OMNIGLOT - 5-Shot, 5-way Hyperbolic ProtoNet Accuracy 99.4 # 14

Methods


No methods listed for this paper. Add relevant methods here