The task for this bachelor practical was to train a convolutional neural network on the Caltech-101 dataset.
My learnings:
- improved knowledge of technologies and concepts (Python, Neural Networks, Machine Learning)
- got to know new technologies and concepts (Keras, Theano, Batch Normalization)
The full report for the bachelor practical can be found here: report.pdf
To get a local copy of the code, clone it using git:
git clone https://github.com/marcuniq/ini_caltech101.git
cd ini_caltech101
Make sure you have the bleeding edge version of Theano, or run
pip install --upgrade --no-deps git+git://github.com/Theano/Theano.git
Next, install the package. Use 'develop' instead of 'install' if you consider changing package code
python setup.py develop
Run train.sh (sets proper theano env flags), which downloads and untars the 'img-gen-resized' dataset, then starts training.
./train.sh