Exploiting Web Images for Fine-Grained Visual Recognition by Eliminating Noisy Samples and Utilizing Hard Ones

TL;DR

This paper presents a method to improve fine-grained visual recognition by filtering noisy web images and leveraging hard examples during training, leading to superior performance over existing web-supervised approaches.

Contribution

The authors introduce a novel technique to eliminate irrelevant noisy samples and utilize hard examples from web images for more robust fine-grained recognition.

Findings

Outperforms state-of-the-art web-supervised methods on three datasets.

Effectively reduces the impact of noisy and irrelevant web images.

Utilizes hard examples to enhance model training.

Abstract

Labeling objects at a subordinate level typically requires expert knowledge, which is not always available when using random annotators. As such, learning directly from web images for fine-grained recognition has attracted broad attention. However, the presence of label noise and hard examples in web images are two obstacles for training robust fine-grained recognition models. Therefore, in this paper, we propose a novel approach for removing irrelevant samples from real-world web images during training, while employing useful hard examples to update the network. Thus, our approach can alleviate the harmful effects of irrelevant noisy web images and hard examples to achieve better performance. Extensive experiments on three commonly used fine-grained datasets demonstrate that our approach is far superior to current state-of-the-art web-supervised methods.