My attempt of the Hateful Meme Detection Challenge:

• Task A is an exploratory data analysis of the objects detected in the memes and creates their catalog which also includes the race, sex and emotion of the humans detected
• Task B is an attempt to analyse the hindering impact of the captions in object detection in the images and uses different methods- TELEA inpainting, NS inpainting and cropping to remove the captions for a comprehensive analysis
• Task C uses 3 models- intermediate fusion of CLIP embeddings, RNN model and an MLP model which use BERT embeddings
• Task D is an analysis of whether NLP models are able to predict hatefulness by exclusion of the image modality

All results are presented in Analysis.pdf