TileLayer

Fix the MVNLayer tests so they actually test what they claim.

MVNLayer fixes: sum_multiplier_ sized correctly; backward gradient calculation.

Gradient calculation per analysis of seanbell, found here:
BVLC#1938

Fixes according to review comments.

Draw Deconvolution layers like Convolution layers

Fix MVNLayer

…==1 in SPPLayer

also, do nothing in SPPLayer Reshape if already reshaped once and bottom size unchanged

Fix SPPLayer top blob num and address `pyramid_height_ == 1`

Give the python layer parameter/weight blobs.

…rning

Fix EmbedLayer compiler warning for unused variable.

Previously, the prefetch GPU -> top GPU and prefetch CPU -> prefetch GPU
copies were launched concurrently in separate streams, allowing the next
batch to be copied in before the current one is read.

This patch explicitly synchronizes the prefetch -> top copy wrt the
host, preventing the CPU -> GPU from being launched until its
completion.

Fix truncation of value warning

Fix a recently introduced race condition in DataLayer

Compute backward for negative lr_mult

Replaces CAffe_POSTFIX -> Caffe_POSTFIX.

This fixes a memory leak by using delete[] rather than plain delete.

Cleanup: Fixup capitalisation of Caffe_POSTFIX.

Commit 4227828 set the default binary format from HDF5 to BINARYPROTO to
fix BVLC#2885. This broke the cifar10 examples which relied on this default.

This commit specifies the snapshot_format explicitly since the rest of the
example relies on this being HDF5.

Fix some doxygen warnings about an undocumented argument in Blob and
incorrect documentation for SoftmaxWithLossLayer::Forward_cpu().

…warnings

Fix up documentation errors.

cifar10: Fix examples by setting snapshot_format.

Show output from convert_imageset tool

Fix memory leak in convert_mnist_siamese_data.

Add extra OpenBLAS include search path

Allow trivial slice/concat

…ble-repeated

NetSpec: don't require lists to specify single-element repeated fields

fix Net debug_info to report learnable params

Fix AccuracyLayerTest for per-class accuracy. Previously in BVLC#2935, it crashes since the test accuracy is nan (0/0) when a class never appear.

Fix AccuracyLayerTest for per-class accuracy.

Fix a typo

[pycaffe] expose all solvers for direct instantiation (although note get_solver)

Minor: missing space in string formatting

Otherwise there seem to be some numerical issues causing BLAS results not exactly same as evaluated results in test code.

Add some documentation on Multi-GPU support

…elayertest

Use EXPECT_NEAR in EltwiseLayer test

Check that the snapshot directory is writeable before starting training

bug in caffe.io.resize_image when applied to images with 2 or >3 channels

OpenCV, LMDB, LevelDB and Snappy are made optional via switches
(USE_OPENCV, USE_LMDB, USE_LEVELDB) available for Make and CMake
builds. Since Snappy is a LevelDB dependency, its use is determined by
USE_LEVELDB. HDF5 is left bundled because it is used for serializing
weights and solverstates.

Separate dependencies for configurable installation

Get back 'USE CPU' print for caffe train

keep old behavior by including leveldb, lmdb, and opencv by default

include IO dependencies by default

ND convolution with im2col

refine format of switch case statement in solver

future-proof version check for BLAS libraries on OS X
fix BVLC#3092

Fix broken DeconvolutionLayer GPU backward caused by ND conv typo

LayerSetUp

This also eliminates the extra copying of bottom's shape.

[build] check xcode command line tools version >= 6

Fix broken conv/deconv reshaping

checking large then small can mask failure since the smaller shape
memory will fit within the larger shape.

Check that output spatial shape varies with input shape while the output
num matches the input num.

[test] Test Reshape more rigorously

[examples] fix link to feature visualization notebook

Allow H5T_INTEGER in HDF5 files

Fix generate_sample_data.py

Fix parse_log.sh against "prefetch queue empty" messages

Disallow PythonLayer in Multi-GPU training

Add a bool flag to record whether a host memory is allocated using malloc or
cudaMallocHost, and free correspondingly using this flag, instead of depending on Caffe::mode(), which is mutable during runtime.

Ensure consistency between memory alloc and free

Fix a typo in the message.

Fix a typo

* shared library
* caffe binary (caffe --version)
* python module (caffe.__version__)

Add argmax_param "axis" to maximise output along the specified axis

488f126

Conflicts:
	Makefile
	Makefile.config.example
	src/caffe/layers/cudnn_conv_layer.cpp
	src/caffe/layers/cudnn_conv_layer.cu
	src/caffe/layers/cudnn_lcn_layer.cpp
	src/caffe/layers/cudnn_lcn_layer.cu

438add5

Conflicts:
	CMakeLists.txt
	cmake/Templates/CaffeConfig.cmake.in

38d6baf

Conflicts:
	src/caffe/layers/cudnn_conv_layer.cpp
	src/caffe/layers/cudnn_conv_layer.cu