Understanding and Evaluating Racial Biases in Image Captioning
This work addresses racial bias in image captioning, which is crucial for accessibility and fairness, but it is incremental as it builds on prior gender bias analyses by adding racial dimensions.
The study investigated racial and intersectional biases in image captioning using manual annotations of perceived gender and skin color on the COCO dataset, finding differences in caption performance, sentiment, and word choice between lighter and darker-skinned people, with modern systems showing greater bias than older ones.
Image captioning is an important task for benchmarking visual reasoning and for enabling accessibility for people with vision impairments. However, as in many machine learning settings, social biases can influence image captioning in undesirable ways. In this work, we study bias propagation pathways within image captioning, focusing specifically on the COCO dataset. Prior work has analyzed gender bias in captions using automatically-derived gender labels; here we examine racial and intersectional biases using manual annotations. Our first contribution is in annotating the perceived gender and skin color of 28,315 of the depicted people after obtaining IRB approval. Using these annotations, we compare racial biases present in both manual and automatically-generated image captions. We demonstrate differences in caption performance, sentiment, and word choice between images of lighter versus darker-skinned people. Further, we find the magnitude of these differences to be greater in modern captioning systems compared to older ones, thus leading to concerns that without proper consideration and mitigation these differences will only become increasingly prevalent. Code and data is available at https://princetonvisualai.github.io/imagecaptioning-bias .