Transparent Human Evaluation for Image Captioning

TL;DR

This paper introduces THumB, a rubric-based human evaluation protocol for image captioning that highlights the limitations of current automatic metrics and emphasizes the importance of comprehensive, transparent human assessments.

Contribution

The paper develops a detailed, rubric-based human evaluation protocol for image captioning, revealing discrepancies between human judgments and automatic metrics.

Findings

Human captions are of higher quality than machine captions in coverage.

CLIPScore correlates better with human judgments than traditional metrics.

Current automatic metrics often underestimate the quality of human-generated captions.

Abstract

We establish THumB, a rubric-based human evaluation protocol for image captioning models. Our scoring rubrics and their definitions are carefully developed based on machine- and human-generated captions on the MSCOCO dataset. Each caption is evaluated along two main dimensions in a tradeoff (precision and recall) as well as other aspects that measure the text quality (fluency, conciseness, and inclusive language). Our evaluations demonstrate several critical problems of the current evaluation practice. Human-generated captions show substantially higher quality than machine-generated ones, especially in coverage of salient information (i.e., recall), while most automatic metrics say the opposite. Our rubric-based results reveal that CLIPScore, a recent metric that uses image features, better correlates with human judgments than conventional text-only metrics because it is more sensitive to recall. We hope that this work will promote a more transparent evaluation protocol for image captioning and its automatic metrics.