Are you happy with your logging solution? Would you help us out by taking a 30-second survey? Click here


Caffe: a fast open framework for deep learning.

Star full 4f7b624809470f25b6493d5a7b30d9b9cb905931146e785d67c86ef0c205a402Star full 4f7b624809470f25b6493d5a7b30d9b9cb905931146e785d67c86ef0c205a402Star full 4f7b624809470f25b6493d5a7b30d9b9cb905931146e785d67c86ef0c205a402Star full 4f7b624809470f25b6493d5a7b30d9b9cb905931146e785d67c86ef0c205a402Star blank 374f33e4d622a2930833db3cbea26b5d03dc44961a6ecab0b9e13276d97d6682 (1 ratings)
Rated 4.0 out of 5
Subscribe to updates I use caffe

Statistics on caffe

Number of watchers on Github 23168
Number of open issues 775
Average time to close an issue 2 days
Main language C++
Average time to merge a PR 6 days
Open pull requests 328+
Closed pull requests 181+
Last commit over 1 year ago
Repo Created about 6 years ago
Repo Last Updated over 1 year ago
Size 63.9 MB
Homepage http://caffe.berk...
Organization / Authorbvlc
Latest Release1.0
Page Updated
Do you use caffe? Leave a review!
View open issues (775)
View caffe activity
View on github
Fresh, new opensource launches ๐Ÿš€๐Ÿš€๐Ÿš€
Trendy new open source projects in your inbox! View examples

Subscribe to our mailing list

Evaluating caffe for your project? Score Explanation
Commits Score (?)
Issues & PR Score (?)
What people are saying about caffe Leave a review
newer ML framework


Build Status License

Caffe is a deep learning framework made with expression, speed, and modularity in mind. It is developed by Berkeley AI Research (BAIR)/The Berkeley Vision and Learning Center (BVLC) and community contributors.

Check out the project site for all the details like

and step-by-step examples.

Custom distributions


Join the chat at

Please join the caffe-users group or gitter chat to ask questions and talk about methods and models. Framework development discussions and thorough bug reports are collected on Issues.

Happy brewing!

License and Citation

Caffe is released under the BSD 2-Clause license. The BAIR/BVLC reference models are released for unrestricted use.

Please cite Caffe in your publications if it helps your research:

  Author = {Jia, Yangqing and Shelhamer, Evan and Donahue, Jeff and Karayev, Sergey and Long, Jonathan and Girshick, Ross and Guadarrama, Sergio and Darrell, Trevor},
  Journal = {arXiv preprint arXiv:1408.5093},
  Title = {Caffe: Convolutional Architecture for Fast Feature Embedding},
  Year = {2014}
caffe open issues Ask a question     (View All Issues)
  • almost 3 years I set the channel in deploy.txt =1 because my images are grayscal images. I need to add is_color=false in data_image_param. First of all in the Alexnet train_val.prototxt there is data_param and not data_image_param!!!! and when I add is_color in data_param I get this error "caffe.DataParameter" has no field named "is_color". Can you please help me?
  • almost 3 years I have two identical GPUs in my machine but when fine tune it seems only one of them is used #3
  • almost 3 years make grayscale lmdb problem
  • almost 3 years CUDNN doesn't support the dilated convolution at Lyaer ***
  • almost 3 years matcaffe compilation error- No such "header" file or directory
  • almost 3 years make: *** [.build_release/cuda/src/caffe/layers/detection_output_layer.o] Error 1
  • almost 3 years Ubuntu 16.04 โ€œ/usr/bin/ld: cannot find -llmdbโ€ error in Caffe compilation
  • almost 3 years version `GLIBCXX_3.4.21' not found
  • almost 3 years Zero accuracy, loss not decreasing and layers have the same value for different inputs
  • almost 3 years Issues with compiling caffe with python, undefined reference to `std::__cxx11::โ€ฆ'
  • almost 3 years error while loading shared libraries:
  • almost 3 years bug report: floating point exception when stepsize 0
  • almost 3 years CUDA error after input layer reshape
  • almost 3 years Training weight dump out
  • almost 3 years problem.. /usr/bin/ld: cannot find -lopencv_imgcodecs
  • almost 3 years Caffe Regression: net prediction is much larger than label but the training loss is very small
  • almost 3 years Installing Caffe Windows
  • almost 3 years I couldn't import _caffe, while I could import caffe. With the help of cmake, I couldn't found
  • almost 3 years .so of caffe which wrapper by myself run very slowly in cgo of golang and thrift(muiti threads)?
  • almost 3 years Caffe-opencl for Windows
  • almost 3 years Fine-Tuning 2 Fully Connected Layers (Classification and Regression)
  • almost 3 years Mac 10.12.1 Sierra Xcode 8.1 cmake 1.vecLib 2.numpy 3.anaconda opencv
  • almost 3 years How to modify the model to fit in your demand?
  • almost 3 years I'm new here, thanks for your help!
  • almost 3 years I met a trouble while trying to "import caffe"
  • almost 3 years Caffe Build Error
  • almost 3 years dlopen Symbol not found
  • almost 3 years test_gradient_based_solver.testbin gives memory corruption
  • almost 3 years Memory leaking ?
  • almost 3 years Loss Layer Should Die Loudly for Unsupported Arguments
caffe open pull requests (View All Pulls)
  • Distributed training
  • Allow EuclideanLossLayer to ignore labels
  • Remove useless LevelDB include
  • Add support for windows build
  • Feat transformation scale
  • Triplet loss
  • Add support for multiple labels to be specified in the image_data_layer
  • reduce lmdb mem map size from 1 TB to 128 MB for MNIST data
  • make CaffeFreeHost thread safe
  • Fix OSX El Capitan incompatibility, by adding CUDA_LIB_DIR to @rpath of
  • Precise accuracy calculation.
  • Improve test output averaging.
  • Create solvers for existing nets
  • Dynamic linking bugfix
  • Python/net spec coordinate map and crop computation
  • Tied weights with transpose flag for InnerProduct layer
  • BiasLayer backward speedup: do backprop to the bias with GEMVs [don't merge]
  • Unpooling Layer
  • Add functions to check and grab GPU
  • Enhance memory data layer
  • Refine P2PSync
  • Fix crash when pairing an odd number of devices without P2P (BVLC/github issue #3531)
  • Fixbug #3494 No to_python (by-value) converter found for C++ tโ€ฆ
  • ND Crop layer
  • solver trace feature to trace weights, loss, diffs and much more
  • Add a Restricted Boltzmann Machine layer
  • Simplified CropLayer
  • Update
  • Fix boost_python discovery for distros with different naming scheme
  • Python implementation of the orthonormal and LSUV initializations
  • Only check the dimensions start from start_axis
  • Added automatic installation script for Linux!
  • Add accuracy
  • Don't force datum.label=0 in array_to_datum
  • Add new orthogonal weight filler (initializer)
  • Document Silence layer in layer catalogue
  • Expose net stages to pycaffe
  • [WIP] Allow for pooling over the first blob axis based on a key
  • LMDB map size - double when full
  • Update db_lmdb.cpp
  • EltwiseLayer supports in-place computation for SUM op
  • command-line interface to modelzoo
  • Suppress boost registration warnings in pycaffe
  • Add external library data layer
  • Fix initialization of deconvolution layer parameters in python net_spec interface
  • Fix Blob Offset Function
  • RNN + LSTM Layers
  • Issue with the exp layer when the base is e
  • removing extra ';' when using the DEFINE_ macros - causing a warning โ€ฆ
  • CUB Memory Manager + cuDNN v4 and v5 support
  • Confusion Matrix
  • small bug in
  • Update info about MKL licensing
  • Missing image checking has been updated in Flickr Style data downloader
  • Add makefile option for Anaconda Python3
  • Fix the bug of incorrect gradient when loss_weight set to zero
  • Added check before register ptr to python
  • Expose all netstate options (for all-in-one nets)
  • upgrading InfogainLoss layer
  • Add Dockerfile from centos7 image
  • add in sudo make uninstall for cmake
  • Wasserstein loss layer
  • mnist get data script bug fix
  • Tests: fix incorrect order of layer_param.set_phase() and DropoutLayer(layer_param)
  • Add clip layer
  • replace open by fopen
  • fix issue in lint.cmake, related to paths, python or variables
  • fix problems in net_surgery.ipynb
  • handle spaces in image file names
  • Fixes
  • add transpose layer and its tests.
  • added pyo files to gitignore
  • Python LMDB Creator
  • add min pooling
  • Windows CMake Build
  • Generalize bilinear filler to for N-D multilinear filler
  • nd convolution and pooling with cuDNN
  • Window data foreground/background nonzero check
  • Fix Makefile CUDA_VERSION extraction on OSX Yosemite
  • CUB Memory Manager header files download
  • More build options (BUILD_EXAMPLES and BUILD_TOOLS)
  • fix flickr finetuning example
  • Create mnist_autoencoder_modified_solver.prototxt
  • Create
  • Create
  • add BLAS option for basic BLAS library
  • Cmake using gnuinstalldirs
  • Exposing HDF5 saving to python
  • Add support for specifying a separator in the image data layer
  • Define preprocessing macro to properly report version on caffe commanโ€ฆ
  • Generalize bilinear interpolation filler to N-D multilinear/multicubic/lanczos filler
  • Added the keyword argument `layer_name` in the method set_input_arrays.
  • convert non-uint8 dtypes to float; refs #2391
  • CUB Memory Manager Integration
  • Update to select different netstate options
  • Fix various documentation typos.
  • Faster R-CNN (work from Ross Girshick)
  • Bug fix for Concat layer in computing offsets for Crop
  • Fix pyCaffe installation on Cmake for MacOS
  • Add Equivalent Pooling Layer
  • Add pruning possibilities at inner_product_layer
  • Set channels, height, and width for encoded images
  • fix sign-compare warning
  • fix the pooling output reshape bug in pooling_layer.cpp
  • Locally connected layer from raingo #3068
  • Fix vecLib search order
  • Examples correction
  • Added PyForwardFromTo and PyBackwardFromTo to Net, for releasing GIL โ€ฆ
  • Fix for Bug# 140953 -โ€ฆ
  • add set_random_seed to the python interface
  • Add clear_param_diffs to the python net interface
  • add layer_dict to the python interface
  • add blob option to set_input_arrays
  • Expand the net.input list to all input layers
  • improve top_names and bottom_names in pycaffe
  • Improve python interface for the solver
  • Fix glog upstream autoconf for Ubuntu 16.04
  • Fixed data/label dimensions when exporting batch containing one "element"
  • Add example for image segmentation.
  • Fix Python installation with CMake install target
  • Windows Python Layer Fix #3915
  • Mass support
  • Introduce a virtual buffer pointer mapping mechanism for OCL memory obj
  • Remove Python.h include since Boost Python already does it
  • [Python] Create solvers/nets without proto files.
  • Add phase support for draw net
  • fix
  • Caffe dist changes
  • net_spec: handle deconvolution layers
  • net_spec: allow lists as tops
  • add default value to rms_decay and fix documentation
  • Kdistil
  • fix the bug of some gpus are blocked during a training iteration
  • Add SetMinLogLevel for setting glog log level.
  • squeeze arrays of output probability
  • Extra interfaces to use caffe operator from mxnet
  • Fix search for Atlas on arch.
  • Mac build
  • CMake: link with ${HDF5_HL_LIBRARIES}
  • Fix
  • add cyclic operations
  • Add random resize to Data Transformer
  • add double-margin contrastive loss layer
  • Fixing relative import for caffe module
  • Improve performance by creating in_place versions of key vector functions
  • Convert examples to an all-in-one networks
  • [windows] Fix missing %(DisableSpecificWarnings) in extract_features.vcxproj
  • fix layerSetUp of scale_layer to not add bias blob when already present
  • Weight sharing bug fixed
  • Correct a minor mistake in statements
  • Recurrent rebase cleanup
  • rc3 fix for mac-build
  • Import bash completion script for caffe from Debian Package.
  • Import manpage caffe.1 from Debian package.
  • Dynamic masking capability for real-time data augmentation
  • Fix more float comparison precision issue
  • Merge NVIDIA's NCCL multi-GPU, switch it to python
  • Caffe examples correction
  • Copyright update
  • Made load_hd5 check blob dims by default, instead of reshaping.
  • Improvements to the build system
  • Add "plateau" LR policy
  • Merge master into Windows
  • Move root_net_ check in net constructor
  • Update caffe.cpp
  • update the module of the scikit-learn (in example/brewing-logreg)
  • Make sure memory is allocated to only one device, enforce if debug
  • Handle filesystems without lock support. NFS can be setup this way.
  • enhance matcaffe: allow explicitly update parameters
  • weight sample by simply adding a new loss layer
  • Hide implementation of LayerRegistry::CreatorRegistry and SolverRegistry::CreatorRegistry singletons
  • Matlab: CMakeLists.txt: install all Matlab files
  • matcaffe: allow destruction of individual networks and solvers
  • fix comments in matlab classification demo
  • added apt command to install OpenBLAS
  • Avoids missing return values during build.
  • added CrossEntropy loss layer
  • CTC (Connectionist Temporal Classification) Implementation
  • Update image_data_layer.cpp
  • Bilinear filler value corrected when matrix size is odd.
  • Make download prebuilt dependencies compatible with python 3.x
  • Add support for vs2015
  • pytest fix: Files created with NamedTemporary files cannot be opened on Windows
  • Checks inside Xcode for latest OSX SDK
  • add instance norm as an option of BatchNormLayer
  • Domain confusion layers for the ICCV2015 paper "Simultaneous Deep
  • Update CMake files to build Matcaffe on Windows
  • Fail when loss_param is used with EuclidanLoss layer
  • Support multi weights file in test mode
  • Update install_apt.html
  • Added CUDA build on appveyor
  • Caffe-opencl for Windows
  • Ssd
  • fix build on Jetson TK1
  • Fixed Point Caffe
  • upgrading matlab interface by adding new methods to matlab class Net
  • Implement CNN Triplet training.
  • Fix Python net drawing code
  • Fix compatibility with CUDA 2.1
  • Add doxygen comments for the batchnorm params.
  • Cosine Loss Layer
  • add note about BLAS
  • example finetune_flickr assemble_data python3 version
  • Redhat atlas library names
  • Multi-GPU segfault fix
  • OpenCL: some optimizations for Intel Gen Graphics
  • Adding numeric gradient check util for python layers
  • Docker refresh
  • bug fix: ext string shouldn't have '.'
  • Overhaul layer catalogue documentation.
  • Fix parse_log tool for negative time duration if datetime across year boundary
  • Crop layer remove redundance
  • Fix batchnorm layer
  • Use the __restrict__ keyword for the low level vector functions
  • Master
  • Simple fix to avoid unnecessary download
  • Join path using "os.path.join" instead of "+"
  • pycaffe fixes and googlenet oversample method
  • use cv2 to load_image, resize_image
  • docs: add debian installation guide
  • Initial commit to add QSML support to Caffe
  • check leveldb iterator status for snappy format.
  • add READme
  • Adding support for python 3.5 in windows
  • display warning message why pycaffe is not built
  • Fix mkl issue #4836
  • Fix
  • Update
  • Filler compatibility for ND convolution and inner product
  • Standardization in data transformer
  • make web_demo work offline
  • fix error link
  • HDF5Output layer with multiple bottom blobs
  • Feature/ignore euclideanloss
  • data transform parallel excution for batch
  • Fix saving using HDF5 output layer using multiple iterations
  • Remove sdk version from veclib searching path.
  • Port nccl parallelism to windows
  • Obsolete reference to `bool solver` in caffe.proto
  • Fix convert_imageset failed when there are multiple blanks in data, lโ€ฆ
  • Support float and integer input types with OpenCV
  • Adjustable clipping
  • Added support for resolving MATLAB prerequisites
  • Removing the duplicated Reshape in Layer->Forward()
  • Create Momentum.cpp
  • typo: `-gpu` -> `--gpu`
  • Make finding MKL on Windows easier
  • Add transformation to MemoryDataLayer / Rebased from PR #3100
  • Added const getters and fixed constness of existing getters for varioโ€ฆ
  • Suppress skimage warning "The default mode, 'constant', will be changed to 'reflect'"
  • Shape mismatch CHECK logging improvements
  • Support Ctrl-C interruption for Matlab training
  • Improved matlab interfaces for robustness
  • Efficient GPU implementation of ReductionLayer forward pass
  • add support CUDA on Jetson-TX1 sm_53, and Jetson-TX2 sm_62
  • fix a bug when data_ or diff_ is shared
  • libdnn: add spatial convolution implmentation
  • store2hdf5.m improve compatibility to different version of matlab
  • Fix extract_seconds for newline endings on Windows
  • Pyloss example backward bug
  • Added support to use NCCL with python3 and fixed nvml.dll related error under windows
  • Solver now copies training stages to testing networks.
  • Remove Makefile.config build completely.
  • MNIST example for windows fixes
  • Reorder BLAS configuration
  • ๅ…ณไบŽmodified_permutohedral.cpp็š„้—ฎ้ข˜
  • Halide Layer
  • Fix missing CUDA_NVCC_FLAGS & CUDA_HOST_COMPILER at GPU detection time.
  • Minor fix on conv layer doc
  • Add support for netlib reference cblas in the CMake build system.
  • Enable C++11 in cmake
  • Tutorial
  • using of center_loss
  • Net object validation check before deleting
  • Only delete Caffe Solver object if reference is defined
  • Added pydot as a requirement
  • Crfrnn
  • Merge pull request #4 from BVLC/master
  • Docker conda
  • copy data for display to avoid changing parameters in the network
  • Update euclidean_loss_layer.hpp with corrected test
  • Handling destruction of empty Solver objects
  • clear bottom diffs after used as temp memory in softmax/sigmoid loss โ€ฆ
  • Add Matlab scripts- log convert & graph plotting
  • Simplify pip invocation.
  • Fix link issue when using Homebrew OpenCV on OS X.
  • fix matchExt
  • Put the acc_data in a new syncedmemory block
  • fix error: duplicate explicit instantiation issue
  • Improvements for issue #6110: The "weights" parameter is added to solver parameters, "snapshot_prefix" defailt initialization
  • Changed paths in example 00-classification.ipynb for portability/consistency
  • Check on solver's parameters to avoid unexplained core dump
  • fix installation issues with makeconfig example and caffe python nameโ€ฆ
  • fix above to below
  • Fix Mac OS X linking for Intel MKL and add comment for LevelDB/TCMalloc problem on Mac OS X
  • [cmake/build] Fix pyCaffe installation on Cmake for MacOS
  • Update blob.hpp for bug issue #5964
  • Make Caffe buildable on OSX 10.11.6 with cmake and python3
  • Added Swish layer
  • add confusion matrix layer for classification task.
  • spatial conv layer for Xception
  • Hotfix for accuracy interfering with training
  • batch normalization : moving average with different batch size
  • Fix for protobuf 3.x
  • Create
  • add
  • Implement CuDNN-based deconvolution layer and test
  • Adding weighted euclidean loss layer
  • CuDNN 7 grouped convolution
  • Some batch scripts for facilitating running examples under Windows
  • Fix caffe rpath
  • [FIX] GPU Build issues on macOS
  • Incorrect namespace for pycaffe submodule caffe_pb2
  • Pycaffe improvements from OpenCL branch
  • relu_layer speed up
  • Patch c make gflags glog
  • Upgrading SoftmaxWithLoss layer to accept also spatially varying weights
  • Fix MultiGPU hangs when interrupting training
  • Remove legacy tools
  • Fix failure to run on system without GPU in CPU mode
  • Expose necessary interface to guide manual SGD process from Python
  • Make caffe compile with CUDA 9.1
  • Merge pull request #6123 from IlyaOvodov/master -> windows
  • BilinearFiller tests refactored
  • PoolingLayer customizable output shape rounding mode
  • check embed index in non debug mode
  • AdaMax solver
  • By leisu. A better softmax_loss_layer.
caffe questions on Stackoverflow (View All Questions)
  • How to use multi CPU cores to train NNs using caffe and OpenBLAS
  • deep learning - a number of naive questions about caffe
  • How to enforce feature vector representing label probability with Caffe siamese CNN?
  • Setting up caffe on Ubuntu 14.04 but facing errors when running
  • Deep Learning Caffe Save Test output labels
  • Caffe/pyCaffe: set all GPUs
  • Input data in caffe
  • Unknown pooling method when testing caffe with cuda but not cudnn
  • Multiple pretrained networks in Caffe
  • scale the loss value according to "badness" in caffe
  • Building Caffe on Ubuntu: make can't find Boost's include files
  • Undefined reference to leveldb when compiling Caffe
  • caffe/python:raise child_exception, OSError: [Errno 2] No such file or directory
  • I am facing some errors while installing Caffe on Ubuntu 14.04
  • Problems building the future branch of caffe (
  • Caffe framework: How can I solve this error?
  • Unable to import caffe and apollocaffe in python. After importing caffe successfully, apollocaffe throws an error and vice versa
  • How to calculate time spent on every layer in CNN with caffe interface?
  • Caffe - Doing forward pass with multiple input blobs
  • Caffe using hdf5 layer and imagedata input layer together in train_val.proto
  • error installing caffe
  • Python layer can't read hdf5 file in caffe framework
  • Input data amount for Caffe
  • caffe fineutne "The same probability of each class"
  • Caffe crashes after first iteration?
  • Caffe Installetion - Can't Start Compilation
  • Converting a Cuda-Convnet checkpoint to a Caffe model binary protobuf
  • Decrease the level of detail in caffe training output?
  • Caffe - Create deploy.prototxt from train_val.prototxt
  • Caffe not reading all h5 files
caffe list of languages used
caffe latest release notes
1.0 1.0

This release marks the convergence of development into a stable, reference release of the framework and a shift into maintenance mode. Let's review the progress culminating in our 1.0:

  • research: nearly 4,000 citations, usage by award papers at CVPR/ECCV/ICCV, and tutorials at ECCV'14 and CVPR'15
  • industry: adopted by Facebook, NVIDIA, Intel, Sony, Yahoo! Japan, Samsung, Adobe, A9, Siemens, Pinterest, the Embedded Vision Alliance, and more
  • community: 250+ contributors, 15k+ subscribers on github, and 7k+ members of the mailing list
  • development: 10k+ forks, >1 contribution/day on average, and dedicated branches for OpenCL and Windows
  • downloads: 10k+ downloads and updates a month, ~50k unique visitors to the home page every two weeks, and >100k unique downloads of the reference models
  • winner of the ACM MM open source award 2014 and presented as a talk at ICML MLOSS 2015

Thanks for all of your efforts leading us to Caffe 1.0! Your part in development, community, feedback, and framework usage brought us here. As part of 1.0 we will be welcoming collaborators old and new to join as members of the Caffe core.

Stay tuned for the next steps in DIY deep learning with Caffe. As development is never truly done, there's always 1.1!

Now that 1.0 is done, the next generation of the frameworkCaffe2is ready to keep up the progress on DIY deep learning in research and industry. While Caffe 1.0 development will continue with 1.1, Caffe2 is the new framework line for future development led by Yangqing Jia. Although Caffe2 is a departure from the development line of Caffe 1.0, we are planning a migration path for models just as we have future-proofed Caffe models in the past.

Happy brewing, The Caffe Crew


rc5 release candidate 5

This packages up 42 commits by 15 contributors to help hone in on 1.0. Thanks all!

With all releases one should do make clean && make superclean to clear out old materials before compiling the new release.

  • set soversion properly #5296
  • documentation: improved dockerfiles and usage notes #5153, links and fixes #5227
  • build: groom cmake build #4609, find veclib more reliably on mac #5236
  • pycaffe: give Net a layer dictionary #4347
  • matcaffe: destroy individual nets and solvers #4737


  • restore solvers for resuming multi-GPU training #5215
  • draw net helper #5010
rc4 release candidate 4

It's a new year and a new release candidate. This packages up 348 commits by 68 authors. Thanks all!

This is intended to be the last release candidate before 1.0. We hope to catch any lurking issues, improve documentation, and polish the packaging for then.

With all releases one should do make clean && make superclean to clear out old materials before compiling the new release. See all merged PRs since the last release.

  • RNNs + LSTMs #3948
  • layers
    • Parameter layer for learning any bottom #2047
    • Crop layer for aligning coordinate maps for FCNs #3570
    • Tied weights with transpose for InnerProduct layer #3612
    • Batch Norm docs, numerics, and robust proto def #4704 #5184
    • Sigmoid Cross Entropy Loss on GPU #4908 and with ignore #4986
  • pycaffe
    • solver callbacks #3020
    • net spec coordinate mapping and cropping for FCNs #3613
    • N-D blob interface #3703
    • python3 compatibility by six #3716
    • dictionary-style net spec #3747
    • Python layer can have phase #3995
  • Docker image #3518
  • expose all NetState options for all-in-one nets #3863
  • force backprop on or off by propagate_down #3942
  • cuDNN v5 #4159
  • multi-GPU parallelism through NCCL + multi-GPU python interface #4563


  • Net upgrade tools catch mixed versions, handle input fields, and log outputs #3755
  • Exp layer for base e and shift != 0 #3937
  • Crop layer checks only the crop dimensions it should #3993


  • cuDNN compatibility is now at v5 + v4 and cuDNN v3 and earlier are not supported
  • NCCL is now required for multi-GPU operation

As a reminder the OpenCL and Windows branches continue to make progress with the community leadership of Fabian Tschopp and Guillaume Dumont resp.


More projects by BVLC View all
Other projects in C++