Authors:
(1) Hyeongjun Kwon, Yonsei University;
(2) Jinhyun Jang, Yonsei University;
(3) Jin Kim, Yonsei University;
(4) Kwonyoung Kim, Yonsei University;
(5) Kwanghoon Sohn, Yonsei University and Korea Institute of Science and Technology (KIST).
Table of Links
4. Method
4.2. Probabilistic hierarchy tree
4.3. Visual hierarchy decomposition
4.4. Learning hierarchy in hyperbolic space
4.5. Visual hierarchy encoding
5. Experiments and 5.1. Image classification
5.2. Object detection and Instance segmentation
6. Ablation studies and discussion
In this document, we include supplementary materials for “Improving Visual Recognition with Hyperbolical Visual Hierarchy Mapping”. We first provide more concrete implementation details (Sec. A), a theoretical baseline (Sec. B), and additional experimental results (Sec. C). Finally, we visualize more visual hierarchy trees from the selected images to provide solid evidence of the proposed method (Sec. D).
A. Network Architecture
A.1. Classification
A.2. Dense prediction.
B. Theoretical Baseline
B.1. Mixture of Gaussians
![Table 6. Performance comparisons between full-training and fine-tuning across various DNNs on the ImageNet-1K dataset [36].](https://hackernoon.imgix.net/images/fWZa4tUiBGemnqQfBGgCPf9594N2-80931zo.png?auto=format&fit=max&w=1200)
This paper is available on arxiv under CC BY 4.0 DEED license.
