Lasse Reedtz

Ai, Neural Nets, Machine Learning



Master Thesis:
Fast convolutional deep learning for image segmentation

My master thesis, written at DTU with Ole Winther and Anders Lindbo Larsen as supervisors, explored the implementation of a cutting edge method for image segmentation using DCNNs on an Nvidia GPU. This was accomplished using CUDArray and Deeppy and was applied to the ISBI Challenge: Segmentation of neuronal structures in EM stacks.


Deeppy and Cudarray

Deeppy and Cudarray are two parts of a python library for Deep neural networks, developed by Anders Lindbo Larsen and me. The library runs on both CPU and GPU and can be run on the GPU using cuDNN or without. The library’s can be found on github and are still under development.
Links: Deeppy And Cudarray