4 tasks with just one network(!) through embeddings

Paras Chopra, managed to perform the following tasks with just one single model:

  • Image caption generator: given an image, generate a caption for it
  • Similar words generator: given a word, find other words similar to it
  • Visually similar image search: given an image, find the most similar images to it
  • Search by describing the content of images: search an image that has contents as described by a textual description

I believe there is potential in multi-use networks especially if the model training is computationally expensive. Read more:

Medium post: https://towardsdatascience.com/one-neural-network-many-uses-image-captioning-image-search-similar-image-and-words-in-one-model-1e22080ce73d