After extensive research on the internet, we found no efficient detector that could identify and flag content containing sexual harassment, sexual abuse, or sexual violence. We observed that none of the state-of-the-art detectors consider sexual harassment or sexual violence in images or videos as offensive content. Thus, our motivation is that existing systems should include this important category of offensive content as an offensive content category. This dataset can be used for machine learning research to build a detector that flags such content as well as identifies the victim and perpetrator in the scenario. With this dataset, we aim to encourage research in this aspect.
The dataset consists of 254 images of scenes from several Bollywood movies depicting sexual harassment, sexual violence and sexual abuse. The images are tagged with either of the three categories mentioned. The images are also annotated with 3 classes: victim, perpetrator and unwanted touching.
The link to the dataset can be found in the file 'dataset.txt'.