CIDEr: Consensus-based image description evaluation

TLDR

Automatically describing an image with a sentence is a long‑standing challenge, and recent advances in object detection, attribute classification, and action recognition have renewed interest, yet evaluating description quality remains difficult. The authors propose a novel evaluation paradigm for image descriptions that relies on human consensus. The paradigm comprises a triplet‑based annotation method, a consensus‑capturing automated metric, and two datasets—PASCAL‑50S and ABSTRACT‑50S—each containing 50 sentences per image. The metric better aligns with human consensus than existing metrics, the authors benchmarked five state‑of‑the‑art image description methods, and they released CIDEr‑D on the MS COCO evaluation server for systematic comparison.

Abstract

Automatically describing an image with a sentence is a long-standing challenge in computer vision and natural language processing. Due to recent progress in object detection, attribute classification, action recognition, etc., there is renewed interest in this area. However, evaluating the quality of descriptions has proven to be challenging. We propose a novel paradigm for evaluating image descriptions that uses human consensus. This paradigm consists of three main parts: a new triplet-based method of collecting human annotations to measure consensus, a new automated metric that captures consensus, and two new datasets: PASCAL-50S and ABSTRACT-50S that contain 50 sentences describing each image. Our simple metric captures human judgment of consensus better than existing metrics across sentences generated by various sources. We also evaluate five state-of-the-art image description approaches using this new protocol and provide a benchmark for future comparisons. A version of CIDEr named CIDEr-D is available as a part of MS COCO evaluation server to enable systematic evaluation and benchmarking.