A Strong and Robust Baseline for Text-Image Matching

TL;DR

This paper introduces a robust baseline for text-image matching by improving training loss functions and inference techniques, leading to better handling of negatives and hubness issues in high-dimensional spaces.

Contribution

It proposes a kNN-margin loss for more robust training and advocates for Inverted Softmax and Cross-modal Local Scaling during inference to improve matching performance.

Findings

kNN-margin loss effectively utilizes hard negatives and tolerates noise

Inverted Softmax and CSLS significantly improve matching scores

The proposed methods outperform existing baselines in text-image matching tasks

Abstract

We review the current schemes of text-image matching models and propose improvements for both training and inference. First, we empirically show limitations of two popular loss (sum and max-margin loss) widely used in training text-image embeddings and propose a trade-off: a kNN-margin loss which 1) utilizes information from hard negatives and 2) is robust to noise as all $K$-most hardest samples are taken into account, tolerating \emph{pseudo} negatives and outliers. Second, we advocate the use of Inverted Softmax (\textsc{Is}) and Cross-modal Local Scaling (\textsc{Csls}) during inference to mitigate the so-called hubness problem in high-dimensional embedding space, enhancing scores of all metrics by a large margin.