DeepLearning

An open source deep learning framework in C++/CUDA

This website introduces the implemetation details of DeepLearning. So users of DeepLearning can use this website as the documentation of the framework.

When I learn something new, one of my habit is to implement the new idea I just learned if possible. For example, after I learned batch normalization, I implemented the Batchnorm layer in the framework, and wrote down the basic idea as well as the implementation details on this website. So, I also personally use this website to record the knowledge during my way of studying deep learning.

Code on GitHub

Features

CUDA

GPU accelerated for all supported layers. You can compile with my hand-crafted CUDA kernels (easy to modify the underlying implementation) or CuDNN (faster execution speed).

Inference and Training

Supports both the training phase and the inference phase. Provide script to directly port the weights trained in PyTorch into DeepLearning.

Common CNN Structures

The Framework comes with configuration files for well-known CNN structures (e.g. VGG, AlexNet, etc). All the provided structures have been test to provide results close to the original paper. More structures will keep coming.

Deep Reinforcement Learning

Currently under development. Not functioning at present.