Image Clustering using Topic Modelling

About

  • Implemented a deep learning-based system that predicts topics of the images using a topic modeling algorithm and clusters them based on predicted topics.
  • Successfully implemented the Latent Dirichlet Allocation (LDA) algorithm using gensim package. It was used to extract topics from the image captions.
  • Used transfer learning approach to extract image features.
  • Fine-tuned the VGG16 & VGG19 model and trained it on Flickr8k and Microsoft COCO datasets for 100 epochs with batch size 64. I was able to achieve a BLEU score of 0.776 on COCO and 0.682 on Flickr8k datasets.