Stop the Nonconsensual Use of Nude Images in Research

Nudity detection is a task that has been studied by researchers for decades. In order to do this work, researchers need datasets of nude content for training, testing, and benchmarking their nudity detection algorithms. To assemble these datasets, researchers typically scrape images from the internet or use existing datasets of nude images. While this practice is common for assembling datasets for general image-recognition tasks, nude images are particularly sensitive. In addition, the nonconsensual collection and distribution of nude images is a common form of image-based sexual abuse, a category of technology-facilitated sexual violence. In this work, we analyzed 153 papers to investigate the use of nude datasets in Computer Science research. Based on our results, we found that researchers regularly collected non-consensual nudes and practiced harmful research processes. We conclude with giving practical recommendations for future research concerning nude datasets.