Deep convolution networks with caffe

download Deep convolution networks with caffe

of 14

  • date post

    04-Jul-2015
  • Category

    Science

  • view

    329
  • download

    4

Embed Size (px)

description

“Deep neural networks. Tool set. Qick start”

Transcript of Deep convolution networks with caffe

  • 1. .. .. - apratster@gmail.com

2. Deep learning1. deep learning. , , , .[1]2. 3. ?4. Deep convolutional neural networks, CAFFE implementation 3. Deep convolutional neural networksConvNet configuration by Krizhevsky [2] 4. (Features sets:)Convolution Neural Network Architecture Model[3] ? ? 5. : [4] , 6. Pooling pooling [5] . 7. deep learninghttp://deeplearning.net/software_links/Caffe deep convolutional neural network frameworkhttp://caffe.berkeleyvision.orgConvNetJS JS based deep learning frameworkhttp://cs.stanford.edu/people/karpathy/convnetjs/DL4J - Java based deep learning frameworkhttp://deeplearning4j.org/Theano CPU/GPU symbolic expression compiler in pythonhttp://deeplearning.net/software/theanoCuda-Convnet A fast C++/CUDA implementation of convolutional(or more generally, feed-forward) neural networkshttp://code.google.com/p/cuda-convnet/Torch provides a Matlab-like environment for state-of-the-art machinelearning algorithms in luahttp://www.torch.ch/Accord.NET - C# deep learninghttp://accord-framework.net/,tutorial:http://whoopsidaisies.hatenablog.com/entry/2014/08/19/015420 8. .. - apratster@gmail.comCAFFE1. GPU (CUDA) and CPU support2. Caffe can be accelerated by NVIDIA cuDNN3. Python and/or MATLAB wrappers4. Config paradigm vs Coding paradigm. Command line tools.*CPU-only Caffe:Uncomment the CPU_ONLY := 1 flag in Makefile.confighttp://caffe.berkeleyvision.org/installation.html 9. CAFFE :build/tools MNIST http://caffe.berkeleyvision.org/gathered/examples/mnist.htmlcd $CAFFE_ROOT./data/mnist/get_mnist.sh./examples/mnist/create_mnist.shcd $CAFFE_ROOT./examples/mnist/train_lenet.sh 10. ?- databases (LevelDB or LMDB)- directly from memory- from files on disk in HDF5- common image formats.http://symas.com/mdb/ http://leveldb.org/Input dataOutput data-snapshot file with mode-snapshot file with solver stateSolver? Yes, we can continue breacked training from snapshot 11. CAFFECaffe stores and communicates data in 4-dimensional arrays called blobsname: "LogReg"layers {name: "mnist"type: DATAtop: "data"top: "label"data_param {source: "input_leveldb"batch_size: 64}} layers {name: "ip"type: INNER_PRODUCTbottom: "data"top: "ip"inner_product_param {num_output: 2}} layers {name: "loss"type: SOFTMAX_LOSSbottom: "ip"bottom: "label"top: "loss"} 12. Convolutional layerRequired fieldnum_output (c_o): the number of filterskernel_size (or kernel_h and kernel_w): specifies height and width of each filterPooling layerRequiredkernel_size (or kernel_h and kernel_w): specifies height and width of each filterLoss Layers, Activation / Neuron Layers, Data Layers, Common LayersHow to configure?Ready to use models in folder: examples 13. 1. , .2. Caffe GPU.3. , .4. , .5. - 6. C++, Python Mathlab. 14. 1. L. Deng and D. Yu, "Deep Learning: Methods and Applicationshttp://research.microsoft.com/pubs/209355/DeepLearning-NowPublishing-Vol7-SIG-039.pdf2. ConvNet configuration by Krizhevsky et alhttp://books.nips.cc/papers/files/nips25/NIPS2012_0534.pdf3. Efficient mapping of the training of Convolutional Neural Networks to a CUDA-basedcluster http://parse.ele.tue.nl/education/cluster24. http://www.cs.toronto.edu/~ranzato/research/projects.html5. http://www.amolgmahurkar.com/classifySTLusingCNN.html !