for creating docker container with required versions of software to develop classifiers run:
make
This it creates and runs jupyter notebook, which allow to develop and run this code.
This repostitory contains three approaches to classification based on transfer learning.
- Hog descriptor from scikit-image v0.14dev + SVM
- Vgg16 network without last layer from Keras with Tensorflow backend + SVM
- Inception V3 with features from pool3:0 layer from Tensorflow + SVM
The SVM is used from liblinear and libsvm libraries with Python wrappers.
There is also used some data augmentation for inception features with over 10000 new images.