CLIPScore: A Reference-free Evaluation Metric for Image Captioning
About
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.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Image Classification | ImageNet 1k (test) | Top-1 Accuracy27.3 | 798 | |
| Image Captioning Evaluation | Composite | Kendall-c Tau_c57.3 | 92 | |
| Image Captioning Evaluation | Flickr8K Expert (test) | Kendall tau_c53 | 76 | |
| Image Captioning Evaluation | Flickr8k Expert | Kendall Tau-c (tau_c)54.3 | 73 | |
| Image Captioning Evaluation | Pascal-50S (test) | HC64.9 | 66 | |
| Perceptual Quality Assessment | HPE-Bench 1.0 (test) | SRCC0.2446 | 66 | |
| Image Captioning Evaluation | Flickr8K-CF (test) | Kendall tau_b36.4 | 65 | |
| Image Captioning Evaluation | Flickr8K-CF | Kendall-b Correlation (tau_b)36.6 | 62 | |
| Image Captioning Evaluation | Pascal-50S | Mean Score84.6 | 39 | |
| Editing Alignment Assessment | HPE-Bench 1.0 (test) | SRCC0.2099 | 33 |