Monk Skin Tone Scale
The Monk Skin Tone Scale is an open-source, 10-shade scale describing human skin color, developed by Ellis Monk in partnership with Google.[1] It is meant to replace the Fitzpatrick scale in fields such as computer vision research, after an IEEE study found the Fitzpatrick scale to be "poorly predictive of skin tone" and advised it "not be used as such in evaluations of computer vision applications."[2] In particular, the Fitzpatrick scale was found to under-represent darker shades of skin relative to the global human population.
The following table shows the 10 categories of the Monk Skin Tone Scale alongside the six categories of the Fitzpatrick scale, grouped into broad skin tone categories:[3]
| Skin tone group | Monk skin tone scale | Fitzpatrick scale |
|---|---|---|
| Light | 1–3 | I–II |
| Medium | 4–6 | III–IV |
| Dark | 7–10 | V–VI |
Predecessor[edit]
Computer vision researchers initially adopted the Fitzpatrick scale as a metric to evaluate how well a given collection of photos of people sampled the global population.[4] However, the Fitzpatrick scale was developed to predict the risk of skin cancer in lighter-skinned people, and did not initially include darker skin tones at all. Two tones for darker people were later added to the original four tones to make it more inclusive. Despite these improvements, research has found that the Fitzpatrick Skin Tone correlated more with self-reported race than with objective measurements of skin tone,[2] and that computer vision models trained using the Fitzpatrick scale perform poorly on images of people with darker skin.[5]
Use[edit]
The Monk scale includes 10 skin tones. Though other scales (such as those used by cosmetics companies) may include many more shades,[6] Monk claims that 10 tones balances diversity with ease of use, and can be used more consistently across different users than a scale with more tones:
Usually, if you got past 10 or 12 points on these types of scales [and] ask the same person to repeatedly pick out the same tones, the more you increase that scale, the less people are able to do that. Cognitively speaking, it just becomes really hard to accurately and reliably differentiate.[5]
The primary intended application of the scale is in evaluating datasets for training computer vision models. Other proposed applications include increasing the diversity of image search results, so that an image search for "doctor" returns images of doctors with a broad range of skin tones.[5]
Google has cautioned against equating the shades in the scale with race, noting that skin tone can vary widely within race.[7]
The Monk scale is licensed under the Creative Commons Attribution 4.0 International license.[8]
See also[edit]
References[edit]
- ^ Monk, Ellis (2023-05-04). "The Monk Skin Tone Scale". SocArXiv. doi:10.31235/osf.io/pdf4c.
- ^ a b Howard, John J.; Sirotin, Yevgeniy B.; Tipton, Jerry L.; Vemury, Arun R. (2021-10-27). "Reliability and Validity of Image-Based and Self-Reported Skin Phenotype Metrics". IEEE Transactions on Biometrics, Behavior, and Identity Science. 3 (4): 550–560. arXiv:2106.11240. doi:10.1109/TBIOM.2021.3123550. ISSN 2637-6407. S2CID 235490065.
- ^ Heldreth, Courtney M.; Monk, Ellis P.; Clark, Alan T.; Schumann, Candice; Eyee, Xango; Ricco, Susanna (31 March 2024). "Which Skin Tone Measures Are the Most Inclusive? An Investigation of Skin Tone Measures for Artificial Intelligence". ACM Journal on Responsible Computing. 1 (1): 1–21. doi:10.1145/3632120. Retrieved 15 April 2024.
- ^ Hazirbas, Caner; Bitton, Joanna; Dolhansky, Brian; Pan, Jacqueline; Gordo, Albert; Ferrer, Cristian Canton (2021-06-19). "Casual Conversations: A dataset for measuring fairness in AI". 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). IEEE. pp. 2289–2293. doi:10.1109/CVPRW53098.2021.00258. ISBN 978-1-6654-4899-4. S2CID 235691583.
- ^ a b c Vincent, James (2022-05-11). "Google is using a new way to measure skin tones to make search results more inclusive". The Verge. Retrieved 2023-06-10.
- ^ "The Scale". Skin Tone Research @ Google. Retrieved 2023-06-10.
- ^ "Recommended Practices". Skin Tone Research @ Google. Retrieved 2023-06-10.
- ^ "Get Started". Skin Tone Research @ Google. Retrieved 2023-06-10.
