Replace pad_, kernel_size_, stride_ with pad_h_, pad_w_, kernel_size_h_, kernel_size_w_, stride_h_, stride_w_ to support pooling on rectangle regions.

Replace pad_, kernel_size_, stride_ with pad_h_, pad_w_, kernel_size_h_, kernel_size_w_, stride_h_, stride_w_ to support pooling on rectangle regions.

Replace pad_, kernel_size_, stride_ with pad_h_, pad_w_, kernel_size_h_, kernel_size_w_, stride_h_, stride_w_ to support pooling on rectangle regions.

Customize compiler setting in Makefile.config

Rectangular pooling

Don't make clean when running linecount

incrementally

"make warn" to print compilation warnings; "make lint" runs incrementally

…stent)

are logged to the warnings file

Fix bug introduced by warning logs (errors didn't print)

Support CPU only memcpy

* master:
  host materials on dl.caffe.berkeleyvision.org
  fix caffe acm-mm paper link
  [docs] readme
  [docs] got rid of redundant README, updated development instructions
  [docs] reworked index page, got rid of publications page
  point to reference performance from installation, add GTX 770
  acknowledge BVLC PI Trevor Darrell for advising Caffe
  add latest CUDA arch to fix invalid device function errors
  switch language to "related publications"
  add publication section to homepage
  fix caffe paper link -- still hasn't appeared on arxiv yet
  Added top-1 and top-5 accuracy for the caffe networks to docs
  add skeleton of the Caffe publications page
  Update docs on building boost on OSX for the python wrappers
  fix OSX 10.9 homebrew CXX doc
  caffe.Net preprocessing members belong to object, not class
  10.9 install doc formatting

net_surgery example to new format

Conflicts:
	docs/getting_pretrained_models.md
	docs/index.md

Add a test param to test both CPU and GPU (with both float and double
Dtypes).

…st Caffe.

Makefile.config instead of CUSTOM_CXX, as Travis exports CXX
as the compiler env variable name.
-Change TEST_HDFS -> TEST_HXX_SRCS.

-Add (CPU-only) test, lint, warn and parallel (-j 4) to travis CI build.
-Add /usr/local/lib to travis config (seems needed for LMDB).
-Put export in "before_script"; disable clang build -- doesn't work on Linux.
-Cache Ubuntu apt packages.
-Install bc package to hopefully suppress "bc: not found" errors
-Get apt packages before_install as suggested by Travis official docs
-Remove specified email address and IRC notifications (emails are sent
to the committer by default; others can view build results in public
Travis feed, on pull requests, etc.).

exit with a non-zero code -- this fixes that.

continuous integration for build, test, and lint by Travis CI

CI build) to fail

Seed HingeLossLayerTest; bad values can cause test (and therefore Travis build) to fail

built are built; write out full -j flag as --jobs for clarity.

Travis tweaks -- make --keep-going; build on all branches; replace CUSTOM_CXX <eom>

Fix bug that breaks MATLAB 2014a compilation

was causing Travis build to randomly fail if a previous test had set the
mode to GPU (which no test that is run by 'make runtestnogpu' should, so
I guess there's another bug somewhere).

Fix SoftmaxLayerTest: forgot to change this one to use DtypesAndDevices;

blobs_lr in Net::Init

…down_

Conditionally compute parameter gradients

MultiDeviceTest (Caffe:set_mode(Caffe::CPU/GPU) isn't run without this).

…hole name space.

- collect CUDA includes in device_alternate.hpp
- add guards for CUDA code
- move GPU code into cu from cpp
- make CUDA includes and libraries conditional in Makefile
- drop CUDA dependency from travis-ci build

CPU-only build

no GPU, no CUDA, no problem!

Minor Code Cleaning

More namespace cleaning.

cosmetics: add syntax = proto2

This makes the default settings usable with Travis/Ubuntu Python

Have Travis build pycaffe

It doesn't make sense to complain about input not being batch-sized if
it isn't even 4-d in the first place.

An input_scale of zero should produce warnings/errors/infinities rather
than being the same as an input_scale of one. A channel_order of () or 0
should not be equivalent to the default ordering.

…ltiplier_

This will make layer reshaping easier and more uniform, and is
consistent with col_buffer_.

…ultiplier_

Again, this is more uniform, and will simplify the implementation of
Reshape.

This will simplify the implementation of Reshape by making allocation
automatic. There is no need for shared_ptr here, and this is more
uniform with buffers used by other layers.

The everything target builds all components, checks warnings, runs lint,
and runs tests.

Add From/To options for Forward and Backward