Learning Visual N-Grams from Web Data

TL;DR

This paper introduces visual n-gram models trained on web data to improve large-scale image recognition, enabling phrase prediction and zero-shot transfer without extensive manual annotation.

Contribution

It presents a novel approach of training visual n-gram models with new loss functions for phrase prediction and image retrieval from webly supervised data.

Findings

Effective phrase prediction from images

Improved image retrieval using visual n-grams

Successful zero-shot transfer capabilities

Abstract

Real-world image recognition systems need to recognize tens of thousands of classes that constitute a plethora of visual concepts. The traditional approach of annotating thousands of images per class for training is infeasible in such a scenario, prompting the use of webly supervised data. This paper explores the training of image-recognition systems on large numbers of images and associated user comments. In particular, we develop visual n-gram models that can predict arbitrary phrases that are relevant to the content of an image. Our visual n-gram models are feed-forward convolutional networks trained using new loss functions that are inspired by n-gram models commonly used in language modeling. We demonstrate the merits of our models in phrase prediction, phrase-based image retrieval, relating images and captions, and zero-shot transfer.