Computer Vision Projects‎ > ‎

Write_kernel_classifier



Citation

This work is a kernel formulation that predicts visual classifiers in the form given by the representer theorem. We also propose a distributional semantic kernel between text descriptions which is shown to be useful in our setting.

 1- Mohamed Elhoseiny, Ahmed Elgammal, Babak Saleh, "Tell and Predict: Kernel Classifier Prediction for Unseen Visual Classes from Unstructured Text Descriptions“, Arxiv, 2015.


Presentations

 1- Mohamed Elhoseiny, Ahmed Elgammal, Babak Saleh, "Tell and Predict: Kernel Classifier Prediction for Unseen Visual Classes from Unstructured Text Descriptions“, CVPR Workshop on Language and Vision, 2015. link

 2- Mohamed Elhoseiny, Ahmed Elgammal, Babak Saleh, "Visual Classifier Prediction by Distributional Semantic Embedding of Text Descriptions“, EMNLP Workshop on Vision and Language, 2015. link





Code 


Word2vec Model  link
  • We used the word2vec Model trained on Google News word corpus (a pretrained model) https://code.google.com/p/word2vec/. It is available here in binary format. Some time ago, we converted it to a MATLAB Containers.Map that takes a word and return a 300 dimensional vector. You don't need to train a new model if you use this one. 
  • For word tokenization and text processing for the word2vec representation. It is mainly done by the function ReadAllTextfiles.mÂ