Low-shot Visual Recognition by Shrinking and Hallucinating Features

Bharath Hariharan, Ross Girshick
6/9/2016
cs.CV

Official code: facebookresearch/low-shot-shrink-hallucinate (310 stars, last commit 2021-08-30).

Abstract

Low-shot visual learning---the ability to recognize novel object categories from very few examples---is a hallmark of human visual intelligence. Existing machine learning approaches fail to generalize in the same way. To make progress on this foundational problem, we present a low-shot learning benchmark on complex images that mimics challenges faced by recognition systems in the wild. We then propose a) representation regularization techniques, and b) techniques to hallucinate additional training examples for data-starved classes. Together, our methods improve the effectiveness of convolutional networks in low-shot learning, improving the one-shot accuracy on novel classes by 2.3x on the challenging ImageNet dataset.

View on arXivView PDF

Code Implementations(1)

facebookresearch/low-shot-shrink-hallucinateOfficial100%

Presenting Low-shot Visual Recognition by Shrinking and Hallucinating Features

31069Shell, PythonMay 25, 20175 years agoNOASSERTION

Similar Papers

The Devil is in the Few Shots: Iterative Visual Knowledge Completion for Few-shot Learning

88% match1 repo132024

What Makes for Effective Few-shot Point Cloud Classification?

87% match1 repo172023

Semantic Prompt for Few-Shot Image Recognition

86% match1 repo442023

Creativity Inspired Zero-Shot Learning

86% match1 repo322019

Not All Features Matter: Enhancing Few-shot CLIP with Adaptive Prior Refinement

86% match1 repo1502023

ImagineFSL: Self-Supervised Pretraining Matters on Imagined Base Set for VLM-based Few-shot Learning

86% match1 repo292025

Cite this paper

@article{hariharan2016lowshot,
  title  = {Low-shot Visual Recognition by Shrinking and Hallucinating Features},
  author = {Bharath Hariharan and Ross Girshick},
  year   = {2016},
  eprint = {1606.02819},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1606.02819}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1606.02819.svg)](https://codeofpaper.com/paper/1606.02819)

Discussion