VQA Demo

Updated to work with Keras 2.0 and TF 1.2 and Spacy 2.0 This code is meant for education thus focus is on simplicity and not speed.

This is a simple Demo of Visual Question answering which uses pretrained models (see models/CNN and models/VQA) to answer a given question about the given image.

Dependency

1. Keras version 2.0+

   • Modular deep learning library based on python

2. Tensorflow 1.2+ (Might also work with Theano. I have not tested Theano after the recent commit, use commit 0f89007 for Theano)

3. scikit-learn

   • Quintessential machine library for python

4. Spacy version 2.0+

   • Used to load Glove vectors (word2vec)
   • To upgrade & install Glove Vectors
     • python -m spacy download en_vectors_web_lg

5. OpenCV

   • OpenCV is used only to resize the image and change the color channels,
   • You may use other libraries as long as you can pass a 224x224 BGR Image (NOTE: BGR and not RGB)

6. VGG 16 Pretrained Weights

   • Please download the weights file vgg16_weights.h5

Usage

python demo.py -image_file_name path_to_file -question "Question to be asked"

e.g

python demo.py -image_file_name test.jpg -question "Is there a man in the picture?"

if you have prefer to use Theano backend and if you have GPU you may want to run like this

THEANO_FLAGS='floatX=float32,device=gpu0,lib.cnmem=1,mode=FAST_RUN' python demo.py -image_file_name test.jpg -question "What vechile is in the picture?"

Expected Output : 095.2 % train 00.67 % subway 00.54 % mcdonald's 00.38 % bus 00.33 % train station

Runtime

• GPU (Titan X) Theano optimizer=fast_run : 51.3 seconds
• GPU (Titan X) Theano optimizer=fast_compile : 47.5 seconds
• CPU (i7-5820K CPU @ 3.30GHz : 35.9 seconds (Is this strange or not ?)

iPython Notebook

Jupyter/iPython Notebook has been provided with more examples and interactive tutorial. https://github.com/iamaaditya/VQA_Demo/blob/master/Visual_Question_Answering_Demo_in_python_notebook.ipynb

NOTE: See the comments on demo.py for more information on the model and methods

VQA Training

• See the repo https://github.com/iamaaditya/VQA_Keras to learn how to train new models