Caption Enriched Samples for Improving Hateful Memes Detection (2109.10649v1)

Abstract: The recently introduced hateful meme challenge demonstrates the difficulty of determining whether a meme is hateful or not. Specifically, both unimodal LLMs and multimodal vision-LLMs cannot reach the human level of performance. Motivated by the need to model the contrast between the image content and the overlayed text, we suggest applying an off-the-shelf image captioning tool in order to capture the first. We demonstrate that the incorporation of such automatic captions during fine-tuning improves the results for various unimodal and multimodal models. Moreover, in the unimodal case, continuing the pre-training of LLMs on augmented and original caption pairs, is highly beneficial to the classification accuracy.