Field-Guide-Inspired Zero-Shot Learning

TL;DR

This paper presents a novel zero-shot learning approach inspired by field guides, where models interactively select the most informative attributes to reduce annotation effort while maintaining high classification accuracy.

Contribution

Introduces an interactive, field-guide-inspired zero-shot annotation method that minimizes attribute annotation costs without sacrificing performance.

Findings

Achieves comparable accuracy to fully annotated models with fewer annotations.

Reduces expert annotation time significantly.

Effective on benchmarks like CUB, SUN, and AWA2.

Abstract

Modern recognition systems require large amounts of supervision to achieve accuracy. Adapting to new domains requires significant data from experts, which is onerous and can become too expensive. Zero-shot learning requires an annotated set of attributes for a novel category. Annotating the full set of attributes for a novel category proves to be a tedious and expensive task in deployment. This is especially the case when the recognition domain is an expert domain. We introduce a new field-guide-inspired approach to zero-shot annotation where the learner model interactively asks for the most useful attributes that define a class. We evaluate our method on classification benchmarks with attribute annotations like CUB, SUN, and AWA2 and show that our model achieves the performance of a model with full annotations at the cost of a significantly fewer number of annotations. Since the time of experts is precious, decreasing annotation cost can be very valuable for real-world deployment.