This is a code for ear segmentation/detection assignment, which was a part of Image based biometry course at Faculty of Computer and Information Science, University of Ljubljana.
The assignment was implemented on AWE dataset [1]. I compared Viola-Jones detector with Mask R-CNN using ResNet50 and MobileNetV2 as backbones. The starting point for my CNN was [2]. I also used different image preprocessing techniques.
I obtained the following results:
[1] http://ears.fri.uni-lj.si/datasets.html#awe-full
[2] https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html