This is a repository for a course project as well as student research work at KTH.
(The template of our report is just a course requirement, not any indication of oriented conference XD)
Demo, a simple demo, { raw images -> preprocessing (segmenting, cropping) -> large scale classifier for classes -> small scale classifier for detecting structures -> show results }
Networks, test code for all models we tried, including VGG16(with FC retrained and with SVM as final classifier), MobileNet and a shallow convolutional network for comparison.
- fa_svm.py file is still under modification, it will not work on your computer.
- since the image dataset is too large, they are not here. and they are non-public too. if you know us and have the right to use the original dataset please send email for our modified one.
Preprocessing, a script for augmenting images, a script for simply turning images into grayscale and 3 channels, a script for splitting the data randomly.
Our VGG16 reference code is from:
(https://github.com/Arsey/keras-transfer-learning-for-oxford102)
The MobileNet reference code is from:
(https://www.tensorflow.org/hub/tutorials/image_retraining)