What question did this study set out to answer?

This research aims to improve object recognition models in scenarios with limited training data, particularly addressing the challenges posed by long-tailed distribution.

June 1, 2010

Comparative object similarity for improved recognition with few or no examples

Key Points

This research aims to improve object recognition models in scenarios with limited training data, particularly addressing the challenges posed by long-tailed distribution.
Developed a regularized kernel machine algorithm incorporating category dependent similarity regularization.
Conducted experiments on hundreds of object categories to validate the model effectiveness.
Analyzed responses to examples from similar and dissimilar categories.
The proposed method showed significant improvement in recognition accuracy, especially for categories without training examples.
Responses were notably stronger for similar categories over dissimilar ones, indicating effective model training.

Abstract

Learning models for recognizing objects with few or no training examples is important, due to the intrinsic long-tailed distribution of objects in the real world. In this paper, we propose an approach to use comparative object similarity. The key insight is that: given a set of object categories which are similar and a set of categories which are dissimilar, a good object model should respond more strongly to examples from similar categories than to examples from dissimilar categories. We develop a regularized kernel machine algorithm to use this category dependent similarity regularization. Our experiments on hundreds of categories show that our method can make significant improvement, especially for categories with no examples.

KI fragen

Bookmark