caffe computer-vision convolutional-neural-network convolutional-neural-networks cuda cudnn deep-learning deep-neural-networks deepflow gpu-computing image-processing machine-learning machine-learning-library mnist neural-networks tensorflow

Deepflow

A node based deep neural network library on top of CuDNN

Status

A work in progress. Not mature enough for a release but cool enough for poking around.

Features

Node base like Tensorflow
Multiple execution phases per graph
Custom solver per variable
Live weight/data display
Create C++ code from Deepflow model

Requirments

NVIDIA Graphics Card
GPU Only

Dependencies

Working Examples

face_dcgan: A least-square generative adversarial network.

face_ac: Deep auto-encoder on 128x128 CelebA dataset.

face_vae: Deep variational auto-encoder on 128x128 CelebA dataset.

mnist_ac: Deep auto-encoder on MNIST dataset.

mnist_dcgan: Least-square gan (generative adversarial) on MNIST dataset.

mnist_lenet

Current Nodes

Nodes	Nodes	Nodes	Nodes
data_generator	variable	place_holder	conv2d
image_batch_generator	pooling	convolution_2d	transposed_conv2d
image_reader	add	square	matmult
mnist_reader	subtract	bias_add	dropout
argmax	argmin	reduce_max	reduce_min
reduce_mean	reduce_sum	reduce_absmax	reduce_norm1
reduce_norm2	leaky_relu	sigmoid	relu
tanh	clipped_relu	elu	phaseplexer
random_selector	softmax_loss	euclidean_loss	print
display	psnr	softmax	equal
cast_float	accumulator	batch_normalization	logger
negate	multiplexer	random_selector	restructure
softmax	...

Current Solvers

AdaDelta
Adam
RMSProp
SGD

About

A node based deep neural network library on top of CuDNN

caffe computer-vision convolutional-neural-network convolutional-neural-networks cuda cudnn deep-learning deep-neural-networks deepflow gpu-computing image-processing machine-learning machine-learning-library mnist neural-networks tensorflow

BSD 3-Clause "New" or "Revised" License

Languages

Language:C++ 71.4%Language:Cuda 24.3%Language:TypeScript 2.5%Language:CMake 0.8%Language:JavaScript 0.5%Language:HTML 0.2%Language:Batchfile 0.1%Language:CSS 0.1%Language:C 0.1%