Low-Shot Learning from Imaginary Data - Robotics Institute Carnegie Mellon University

Low-Shot Learning from Imaginary Data

Yuxiong Wang, Ross Girshick, Martial Hebert, and Bharath Hariharan
Conference Paper, Proceedings of (CVPR) Computer Vision and Pattern Recognition, pp. 7278 - 7286, June, 2018

Abstract

Humans can quickly learn new visual concepts, perhaps because they can easily visualize or imagine what novel objects look like from different views. Incorporating this ability to hallucinate novel instances of new concepts might help machine vision systems perform better low-shot learning, i.e., learning concepts from few examples. We present a novel approach to low-shot learning that uses this idea. Our approach builds on recent progress in meta-learning (''learning to learn'') by combining a meta-learner with a ''hallucinator'' that produces additional training examples, and optimizing both models jointly. Our hallucinator can be incorporated into a variety of meta-learners and provides significant gains: up to a 6 point boost in classification accuracy when only a single training example is available, yielding state-of-the-art performance on the challenging ImageNet low-shot classification benchmark.

BibTeX

@conference{Wang-2018-105271,
author = {Yuxiong Wang and Ross Girshick and Martial Hebert and Bharath Hariharan},
title = {Low-Shot Learning from Imaginary Data},
booktitle = {Proceedings of (CVPR) Computer Vision and Pattern Recognition},
year = {2018},
month = {June},
pages = {7278 - 7286},
}