{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/clipscore-a-reference-free-evaluation-metric","title":"CLIPScore: A Reference-free Evaluation Metric for Image Captioning","arxiv_id":"2104.08718","date":"2021-04-18","proceeding":"EMNLP 2021 11","authors":["Jack Hessel","Ari Holtzman","Maxwell Forbes","Ronan Le Bras","Yejin Choi"],"abstract":"Image captioning has conventionally relied on reference-based automatic evaluations, where machine captions are compared against captions written by humans. This is in contrast to the reference-free manner in which humans assess caption quality. In this paper, we report the surprising empirical finding that CLIP (Radford et al., 2021), a cross-modal model pretrained on 400M image+caption pairs from the web, can be used for robust automatic evaluation of image captioning without the need for references. Experiments spanning several corpora demonstrate that our new reference-free metric, CLIPScore, achieves the highest correlation with human judgements, outperforming existing reference-based metrics like CIDEr and SPICE. Information gain experiments demonstrate that CLIPScore, with its tight focus on image-text compatibility, is complementary to existing reference-based metrics that emphasize text-text similarities. Thus, we also present a reference-augmented version, RefCLIPScore, which achieves even higher correlation. Beyond literal description tasks, several case studies reveal domains where CLIPScore performs well (clip-art images, alt-text rating), but also where it is relatively weaker in comparison to reference-based metrics, e.g., news captions that require richer contextual knowledge.","url_abs":"https://arxiv.org/abs/2104.08718v3","url_pdf":"https://arxiv.org/pdf/2104.08718v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"clipscore-a-reference-free-evaluation-metric","repo_url":"https://github.com/jmhessel/clipscore","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"clipscore-a-reference-free-evaluation-metric","repo_url":"https://github.com/jmhessel/pycocoevalcap","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"NOASSERTION"}},{"paper_slug":"clipscore-a-reference-free-evaluation-metric","repo_url":"https://github.com/showlab/loveu-tgve-2023","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":null,"task_name":"Hallucination Pair-wise Detection (1-ref)"},{"task_slug":null,"task_name":"Hallucination Pair-wise Detection (4-ref)"},{"task_slug":"human-judgment-classification","task_name":"Human Judgment Classification"},{"task_slug":"human-judgment-correlation","task_name":"Human Judgment Correlation"},{"task_slug":"image-captioning","task_name":"Image Captioning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/human-judgment-classification-on-pascal-50s","task":"Human Judgment Classification","dataset":"Pascal-50S","model":"RefCLIP-S","rank_in_archive_order":2,"of":3,"metrics":{"Mean Accuracy":"83.1"},"uses_additional_data":false},{"leaderboard":"/sota/human-judgment-classification-on-pascal-50s","task":"Human Judgment Classification","dataset":"Pascal-50S","model":"CLIP-S","rank_in_archive_order":3,"of":3,"metrics":{"Mean Accuracy":"80.7"},"uses_additional_data":false},{"leaderboard":"/sota/human-judgment-correlation-on-flickr8k-cf","task":"Human Judgment Correlation","dataset":"Flickr8k-CF","model":"RefCLIP-S","rank_in_archive_order":2,"of":3,"metrics":{"Kendall's Tau-b":"36.4"},"uses_additional_data":false},{"leaderboard":"/sota/human-judgment-correlation-on-flickr8k-cf","task":"Human Judgment Correlation","dataset":"Flickr8k-CF","model":"CLIP-S","rank_in_archive_order":3,"of":3,"metrics":{"Kendall's Tau-b":"34.4"},"uses_additional_data":false},{"leaderboard":"/sota/human-judgment-correlation-on-flickr8k-expert","task":"Human Judgment Correlation","dataset":"Flickr8k-Expert","model":"RefCLIP-S","rank_in_archive_order":3,"of":4,"metrics":{"Kendall's Tau-c":"53.0"},"uses_additional_data":false},{"leaderboard":"/sota/human-judgment-correlation-on-flickr8k-expert","task":"Human Judgment Correlation","dataset":"Flickr8k-Expert","model":"CLIP-S","rank_in_archive_order":4,"of":4,"metrics":{"Kendall's Tau-c":"51.2"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2104.08718","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}