Exposing and Correcting the Gender Bias in Image Captioning Datasets and Models

By Shruti Bhargava, David A. Forsyth. 2019

59 citations recorded by the source corpus.

The task of image captioning implicitly involves gender identification. However, due to the gender bias in data, gender identification by an image captioning model suffers. Also, the gender-activity bias, owing to the word-by-word prediction, influences other words in the caption prediction, resulting in the well-known problem of label bias. In this work, we investigate gender bias in the COCO captioning dataset and show that it engenders not only from the statistical distribution of genders with contexts but also from the flawed annotation by the human annotators. We look at the issues created by this bias in the trained models. We propose a technique to get rid of the bias by splitting the task into 2 subtasks: gender-neutral image captioning and gender classification. By this decoupling, the gender-context influence can be eradicated. We train the gender-neutral image captioning model, which gives comparable results to a gendered model even when evaluating against a dataset that possesses a similar bias as the training data. Interestingly, the predictions by this model on images with no humans, are also visibly different from the one trained on gendered captions. We train gender classifiers using the available bounding box and mask-based annotations for the person in the image. This allows us to get rid of the context and focus on the person to predict the gender. By substituting the genders into the gender-neutral captions, we get the final gendered predictions. Our predictions achieve similar performance to a model trained with gender, and at the same time are devoid of gender bias. Finally, our main result is that on an anti-stereotypical dataset, our model outperforms a popular image captioning model which is trained with gender.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Safety, Robustness & Interpretability

Citations in the corpus

Language (Technology) is Power: A Critical Survey of “Bias” in NLP (2020, neutral): “…al. (2018); Kiritchenko and Mohammad (2018); Lu et al. (2018); Park et al. (2018); Shen et al. (2018); Zhang et al. (2018); Badjatiya et al. (2019); Bhargava and Forsyth (2019); Cao and Daumé (2019); Cho et al. (2019); Davidson et al. (2019); Dev et al. (2019); Garimella et al. (2019); Garg et…”
Underspecification in Scene Description-to-Depiction Tasks (2022, background): “The strong dependence on the writer, in particular, highlights the socially and culturally subjective nature of image descriptions (van Miltenburg et al., 2017; Bhargava and Forsyth, 2019).”
Mitigating Gender Bias in Captioning Systems (2020, background): “e from the vision part. Mitigating Gender Bias: Few initial attempts have been made to design captioning models to overcome gender biases in datasets. One solution is to break the task into two steps [36]. It ﬁrstly locates and recognizes the person in the image. Then a language model utilizes groun”
DALL-Eval: Probing the Reasoning Skills and Social Biases of Text-to-Image Generative Transformers (2022, background): “[7, 8, 55] examined social biases on the image captioning datasets: MS COCO [36] and LAION-400M.”
DALL-EVAL: Probing the Reasoning Skills and Social Biases of Text-to-Image Generation Models (2022, unknown): “[6, 5, 58, 9, 73, 28, 62, 27] examine social biases in image-text datasets.”
Taxonomizing and Measuring Representational Harms: A Look at Image Tagging (2023, background): “…that the performance of image tagging systems can vary according to the identities of the people depicted in images (Hendricks et al. 2018; Zhao et al. 2017), and revealed a variety of different ways that such systems can perpetuate stereotypes (Schwemmer et al. 2020; Bhargava and Forsyth 2019).”
Stable Bias: Evaluating Societal Representations in Diffusion Models (2023, unknown): “…vision and language models found that societal biases regarding race, gender and appearance propagate to downstream outputs in tasks such as image captioning (e.g., [28, 9]) and image search (e.g., [70, 42]), and has found that such biases are learned by state-of-the-art multimodal models [67, 68].”