Label-Embedding for Image Classification

TLDR

Attributes serve as intermediate representations that enable parameter sharing across classes, which is essential when training data is scarce. The study proposes treating attribute‑based image classification as a label‑embedding problem, embedding each class in the space of attribute vectors. The authors introduce a compatibility function between images and label embeddings, learn its parameters on labeled data to rank correct classes higher, and show that label embedding can incorporate additional information such as class hierarchies or textual descriptions, thereby covering settings from zero‑shot to fully supervised learning. On the Animals With Attributes and Caltech‑UCSD‑Birds datasets, the framework outperforms the Direct Attribute Prediction baseline in zero‑shot learning.

Abstract

Attributes act as intermediate representations that enable parameter sharing between classes, a must when training data is scarce. We propose to view attribute-based image classification as a label-embedding problem: each class is embedded in the space of attribute vectors. We introduce a function that measures the compatibility between an image and a label embedding. The parameters of this function are learned on a training set of labeled samples to ensure that, given an image, the correct classes rank higher than the incorrect ones. Results on the Animals With Attributes and Caltech-UCSD-Birds datasets show that the proposed framework outperforms the standard Direct Attribute Prediction baseline in a zero-shot learning scenario. Label embedding enjoys a built-in ability to leverage alternative sources of information instead of or in addition to attributes, such as, e.g., class hierarchies or textual descriptions. Moreover, label embedding encompasses the whole range of learning settings from zero-shot learning to regular learning with a large number of labeled examples.