Deep Vision for Dummies

• To install from Github do,

pip install git+git://github.com/ajayrfhp/dvd/

• To install from PIP do,

pip install dvd

Installation involves downloading of vgg model into /tmp/. This is a huge file, this takes about 20 minutes if you have good internet speed. Please wait.

Usage :

• Check examples folder for a demo notebook.
  • Included transfer learning for MNIST, achieve 99.2% accuracy with logistic regression using transfer learnt representation as features
• Features as of now
  • Transfer Learning
    • Takes input an image and returns a trasnfer learnt representation of the image.
    • The returned vector can be used as features of image to build classification algorithms using simple ML models like logistic regression, SVM.
    • Get transfer learnt embeddings from Vgg Net. Given an image obtain a dense vector representation.
  • Object Detector
    • Simple object detector
    • More Coming
  • Image to text
    • Coming up
  • Text to image
    • Coming up

This work is possible because of Awesome place to get Vgg Net, by Professor Davi frosard.

The Vgg model code and data used are from Professor Davi Frosard's website.