Evan Shelhamer [Wed, 11 Nov 2015 06:50:28 +0000 (22:50 -0800)]
Merge pull request #3290 from ernest-tg/master
[pycaffe] correct transposition & channel_swap in deprocess
Evan Shelhamer [Wed, 11 Nov 2015 06:48:23 +0000 (22:48 -0800)]
Merge pull request #3294 from panmari/simpler_diff
[example] switch order of two layers for simpler diff with untuned file
Evan Shelhamer [Wed, 11 Nov 2015 06:47:27 +0000 (22:47 -0800)]
Merge pull request #3295 from timmeinhardt/fix_issue_3274
[bug] fix issue #3274 -- shape argmax top carefully
Evan Shelhamer [Wed, 11 Nov 2015 06:44:46 +0000 (22:44 -0800)]
Merge pull request #3310 from gustavla/contrastive-doc-fix
[doc] Fix consistent typo in contrastive loss
Ronghang Hu [Tue, 10 Nov 2015 23:08:47 +0000 (15:08 -0800)]
Merge pull request #3305 from ronghanghu/display-ignore-layer
display 'ignore source layer' when initializing from existing parameters
Ronghang Hu [Tue, 10 Nov 2015 23:08:22 +0000 (15:08 -0800)]
Merge pull request #3300 from BonsaiAI/replace-unistd
Replace unistd functions with cross platform counterparts
Tea [Sat, 7 Nov 2015 06:09:59 +0000 (14:09 +0800)]
Replace unistd functions with cross platform counterparts
Gustav Larsson [Mon, 9 Nov 2015 20:32:37 +0000 (14:32 -0600)]
DOC: Fix consistent typo in contrastive loss
If a pair is similar, it should take the squared distance and not the
distance. This is clearly what the code is doing.
Ronghang Hu [Mon, 9 Nov 2015 18:15:26 +0000 (10:15 -0800)]
Merge pull request #3308 from cassinaj/master
minor typo fix
Jan Issac [Mon, 9 Nov 2015 11:42:56 +0000 (12:42 +0100)]
Merge pull request #1 from cassinaj/cassinaj-minor-typo-fix
minor typo fix
panmari [Fri, 6 Nov 2015 11:53:16 +0000 (12:53 +0100)]
Switched order of two layers for simpler diff with untuned file
Untuned file is in models/bvlc_reference_caffenet/train_val.prototxt.
Ronghang Hu [Sun, 8 Nov 2015 19:51:24 +0000 (11:51 -0800)]
Merge pull request #3303 from eelstork/patch-1
GetDB must return a value
Ronghang Hu [Sun, 8 Nov 2015 19:20:32 +0000 (11:20 -0800)]
display 'ignore source layer' when initializing from existing parameters
This helps in the case to see which layer is initialized from existing parameters, and which layer is ignored. This helps identify the cases where the user types a error mismatch layer name.
T.E.A de Souza [Sun, 8 Nov 2015 10:50:29 +0000 (18:50 +0800)]
GetDB must return a value
As noted by @danst18, when USE_LEVELDB and USE_LMDB are disabled, a compiler error is issued since GetDB no longer returns a value.
At runtime a fatal error would be issued anyways. However to help users who don't need a DB backend, NULL should be returned here.
Tim Meinhardt [Fri, 6 Nov 2015 13:51:46 +0000 (14:51 +0100)]
Fix ArgMaxLayer::Reshape for any num of bottom axes
ernest-tg [Thu, 5 Nov 2015 14:47:28 +0000 (15:47 +0100)]
Correct transposition & channel_swap in deprocess
The deprocess( ) function should invert the preprocess( ) function, however it only worked when the permutation of your channel_swap is of order 2 and the permutation of your transpose were of order 3. This is usually the case, which is why this bug went unnoticed for a long time.
To reproduce it (on former version), try to preprocess and then deprocess with
transformer.set_transpose('data', (0,2,1)) (or (1,0,2) or (2,1,0))
Or with
transformer.set_channel_swap('data', (2,0,1)) (or (1,2,0) )
Indeed, we had
L152 (in preprocess) caffe_in = caffe_in[channel_swap, :, :]
L181 (in deprocess) decaf_in = decaf_in[channel_swap, :, :]
So we applied [channel_swap,:,:] twice to the initial data => not always the identity
L154 (in preprocess) caffe_in = caffe_in.transpose(transpose)
L183 (in deprocess) decaf_in = decaf_in.transpose([transpose[t] for t in transpose])
The transposition [transpose[t] for t in transpose] is (tranpsose)² so we applied transpose[t] three times which is not always the identity.
Jon Long [Thu, 5 Nov 2015 08:27:56 +0000 (00:27 -0800)]
Merge pull request #3287 from longjon/travis-install-whitespace
Fix whitespace in travis_install.sh
Jonathan L Long [Thu, 5 Nov 2015 07:54:41 +0000 (23:54 -0800)]
[style] fix whitespace in travis_install.sh
Jon Long [Thu, 5 Nov 2015 07:49:15 +0000 (23:49 -0800)]
Merge pull request #3286 from longjon/fix-py3-travis-boost
[travis] fix boost/python3 conda conflict
Jonathan L Long [Thu, 5 Nov 2015 06:10:25 +0000 (22:10 -0800)]
[travis] fix boost/python3 conda conflict
Jon Long [Thu, 5 Nov 2015 05:32:16 +0000 (21:32 -0800)]
Merge pull request #3275 from darxriggs/master
fix detect.py (invalid model path)
Evan Shelhamer [Tue, 3 Nov 2015 22:52:01 +0000 (14:52 -0800)]
Merge pull request #3276 from lukeyeager/bvlc/fix-travis-cmake-download
TravisCI: wget cmake with --no-check-certificate
Luke Yeager [Tue, 3 Nov 2015 22:42:24 +0000 (14:42 -0800)]
TravisCI: wget cmake with --no-check-certificate
```
--2015-11-03 22:31:11-- http://www.cmake.org/files/v3.2/cmake-3.2.3-Linux-x86_64.sh
Resolving www.cmake.org (www.cmake.org)... 66.194.253.19
Connecting to www.cmake.org (www.cmake.org)|66.194.253.19|:80... connected.
HTTP request sent, awaiting response... 301 Moved Permanently
Location: http://cmake.org/files/v3.2/cmake-3.2.3-Linux-x86_64.sh [following]
--2015-11-03 22:31:11-- http://cmake.org/files/v3.2/cmake-3.2.3-Linux-x86_64.sh
Resolving cmake.org (cmake.org)... 66.194.253.19
Connecting to cmake.org (cmake.org)|66.194.253.19|:80... connected.
HTTP request sent, awaiting response... 301 Moved Permanently
Location: https://cmake.org/files/v3.2/cmake-3.2.3-Linux-x86_64.sh [following]
--2015-11-03 22:31:11-- https://cmake.org/files/v3.2/cmake-3.2.3-Linux-x86_64.sh
Connecting to cmake.org (cmake.org)|66.194.253.19|:443... connected.
ERROR: no certificate subject alternative name matches
requested host name `cmake.org'.
To connect to cmake.org insecurely, use `--no-check-certificate'.
```
René Scheibe [Tue, 3 Nov 2015 18:27:07 +0000 (19:27 +0100)]
fix detect.py (invalid model path)
Evan Shelhamer [Sat, 31 Oct 2015 06:08:12 +0000 (23:08 -0700)]
Merge pull request #3082 from gustavla/pycaffe-snapshot
Expose `Solver::Snapshot` to pycaffe
Ronghang Hu [Wed, 28 Oct 2015 17:19:33 +0000 (10:19 -0700)]
Merge pull request #3254 from ronghanghu/cudnn3_accum_grad
CuDNNConvolutionLayer accumulate gradients
Ronghang Hu [Wed, 28 Oct 2015 05:49:28 +0000 (22:49 -0700)]
CuDNNConvolutionLayer accumulate gradients
Evan Shelhamer [Sun, 25 Oct 2015 00:05:12 +0000 (17:05 -0700)]
Merge pull request #3221 from BonsaiAI/cmake-forward-compatibility
Endorse CMP0046, CMP0054
Jeff Donahue [Fri, 23 Oct 2015 01:33:07 +0000 (18:33 -0700)]
Merge pull request #3229 from cdoersch/batchnorm2
Yet another batch normalization PR
Ronghang Hu [Thu, 22 Oct 2015 22:17:26 +0000 (15:17 -0700)]
Merge pull request #3186 from lionleaf/makefile_opencv_imgcodecs
Add opencv_imgcodecs to library path in Makefile
Andreas Løve Selvik [Tue, 13 Oct 2015 14:32:52 +0000 (16:32 +0200)]
Add opencv_imgcodecs to library path in Makefile
Project does not compile without opencv_imgcodecs in the library path if
you're using OpenCV 3.
This introduces a OPENCV_VERSION flag in Makefile.config that includes
the library if set to 3. (Trying to include it with OpenCV 2 also
breaks the build)
Ronghang Hu [Thu, 22 Oct 2015 17:11:30 +0000 (10:11 -0700)]
Merge pull request #3234 from rodrigob/patch-1
diff.ndim != 4 is outdated
Rodrigo Benenson [Thu, 22 Oct 2015 16:18:08 +0000 (18:18 +0200)]
diff.ndim != 4 is outdated
this code seems not to apply to the caffe head.
```
if diff.ndim != 4:
raise Exception('{} diff is not 4-d'.format(top))
```
Carl Doersch [Tue, 6 Oct 2015 21:19:59 +0000 (14:19 -0700)]
Cleanup batch norm layer, include global stats computation
Ronghang Hu [Thu, 22 Oct 2015 04:22:24 +0000 (21:22 -0700)]
Merge pull request #3226 from kkhoot/clean_header
Clean includes
Dmytro Mishkin [Wed, 25 Feb 2015 15:00:22 +0000 (17:00 +0200)]
Added batch normalization layer with test and examples
Jon Long [Wed, 21 Oct 2015 02:17:06 +0000 (19:17 -0700)]
Merge pull request #3088 from lukeyeager/bvlc/lmdb-nolock
Open LMDB files with MDB_NOLOCK if no write access
Kang Kim [Sun, 18 Oct 2015 15:58:55 +0000 (00:58 +0900)]
Clean redundant/unnecessary headers
Kang Kim [Sun, 18 Oct 2015 07:52:19 +0000 (16:52 +0900)]
Move HDF5 defines to data_layers header
T.E.A de Souza [Mon, 19 Oct 2015 10:13:57 +0000 (18:13 +0800)]
Endorse CMP0046, CMP0054
Set policies to NEW to silence warnings in CMake 3.02 and later.
Evan Shelhamer [Mon, 19 Oct 2015 22:21:32 +0000 (15:21 -0700)]
Merge pull request #3182 from bchu/tutorial-fix
[example] Fix class label index error in tutorial 00.
Ronghang Hu [Mon, 19 Oct 2015 18:48:19 +0000 (11:48 -0700)]
Merge pull request #3185 from mausoomsarkar/master
PReLU speed up
Evan Shelhamer [Mon, 19 Oct 2015 18:46:14 +0000 (11:46 -0700)]
Merge pull request #3205 from shelhamer/test-layer-type-upgrade
[test] drop bogus OpenCV guard for layer type
Evan Shelhamer [Sat, 17 Oct 2015 00:13:24 +0000 (17:13 -0700)]
[test] drop bogus OpenCV guard for layer type
Evan Shelhamer [Mon, 19 Oct 2015 18:43:59 +0000 (11:43 -0700)]
Merge pull request #3218 from shelhamer/cudnn3-compatible
[docs] mark cuDNN v3 compatible
Evan Shelhamer [Mon, 19 Oct 2015 18:43:36 +0000 (11:43 -0700)]
Merge pull request #3219 from shelhamer/install-caffe-users
[docs] direct installation questions to caffe-users
Evan Shelhamer [Mon, 19 Oct 2015 18:39:29 +0000 (11:39 -0700)]
[docs] cuDNN v3 compatible
Evan Shelhamer [Mon, 19 Oct 2015 18:36:38 +0000 (11:36 -0700)]
installation questions -> caffe-users
Evan Shelhamer [Mon, 19 Oct 2015 17:05:14 +0000 (10:05 -0700)]
Merge pull request #3217 from BonsaiAI/qualify-CUDA-status-message
[build] Qualify messages issued by CMake when CUDA is unavailable
T.E.A de Souza [Mon, 19 Oct 2015 10:19:38 +0000 (18:19 +0800)]
Qualify messages issued by CMake when CUDA is unavailable
Mausoom Sarkar [Tue, 13 Oct 2015 13:05:32 +0000 (18:35 +0530)]
Moved the loop inside PReLUParamBackward to do the reduction inside the kernel
Now PReLU backward is taking the same time as forward
Code cleanup
Removed unnecessary code
Fixed indent
merge if(channed_shared_)
Evan Shelhamer [Sat, 17 Oct 2015 07:26:40 +0000 (00:26 -0700)]
Merge pull request #3116 from ronghanghu/solver-refactor
Solver Refactor: Separate files and Change Solver's Type to String
Evan Shelhamer [Sat, 17 Oct 2015 07:23:48 +0000 (00:23 -0700)]
Merge pull request #3206 from shelhamer/net-init-logging
[style] Clean Up Logging for Net Init
Evan Shelhamer [Sat, 17 Oct 2015 00:32:27 +0000 (17:32 -0700)]
clean up logging for Net init
- condense conditions by `LOG_IF`
- only log memory use once after all tops
Ronghang Hu [Sat, 26 Sep 2015 18:47:32 +0000 (11:47 -0700)]
Update examples and docs
Ronghang Hu [Sat, 26 Sep 2015 18:47:02 +0000 (11:47 -0700)]
Add automatic upgrade for solver type
Ronghang Hu [Fri, 25 Sep 2015 02:40:45 +0000 (19:40 -0700)]
Change solver type to string and provide solver registry
Ronghang Hu [Fri, 25 Sep 2015 00:11:07 +0000 (17:11 -0700)]
Split solver code into one file per solver class
Ronghang Hu [Sat, 17 Oct 2015 05:01:26 +0000 (22:01 -0700)]
Merge pull request #3207 from ronghanghu/test-matcaffe-io
Test reading and writing mean proto in matlab
Ronghang Hu [Sat, 17 Oct 2015 04:19:59 +0000 (21:19 -0700)]
Test reading and writing mean proto in matlab
Evan Shelhamer [Sat, 17 Oct 2015 01:09:00 +0000 (18:09 -0700)]
Merge pull request #3089 from shelhamer/groom-conv
[style] groom im2col + col2im for clarity
Evan Shelhamer [Sat, 17 Oct 2015 01:08:22 +0000 (18:08 -0700)]
Merge pull request #3204 from shelhamer/rearrange-upgrade-helpers
[style] Rearrange Proto Upgrade Helpers
Evan Shelhamer [Sat, 17 Oct 2015 00:59:13 +0000 (17:59 -0700)]
Merge pull request #3190 from DrSleep/master
Fix drawing problems with repeated convolution -- close #3201
Evan Shelhamer [Sat, 17 Oct 2015 00:56:13 +0000 (17:56 -0700)]
Merge pull request #3189 from bchu/caffenet-mirror
Set CaffeNet train_val test mirroring to false
Evan Shelhamer [Fri, 16 Oct 2015 23:33:06 +0000 (16:33 -0700)]
rearrange upgrade helpers
order from general helpers to specific upgrades in chronological order.
Evan Shelhamer [Fri, 16 Oct 2015 03:17:55 +0000 (20:17 -0700)]
Merge pull request #3160 from shelhamer/cudnnV3
Basic cuDNN v3 support
Simon Layton [Wed, 8 Jul 2015 19:35:55 +0000 (15:35 -0400)]
Initial cuDNN v3 support
Jeff Donahue [Thu, 15 Oct 2015 21:16:03 +0000 (14:16 -0700)]
Merge pull request #3200 from lukeyeager/bvlc/fix-blobproto_to_array
Allow old-style shape in blobproto_to_array
Luke Yeager [Thu, 15 Oct 2015 18:03:09 +0000 (11:03 -0700)]
Allow old-style shape in blobproto_to_array
Fixes #3199
Bug introduced in #3170
Vladimir [Wed, 14 Oct 2015 03:00:14 +0000 (12:00 +0900)]
Fixed drawing problems with repeated convolution
Jeff Donahue [Wed, 14 Oct 2015 00:25:09 +0000 (17:25 -0700)]
Merge pull request #2966 from cdoersch/batch_reindex_layer
BatchReindexLayer to shuffle, subsample, and replicate examples in a batch
Brian Chu [Tue, 13 Oct 2015 20:24:42 +0000 (13:24 -0700)]
Set CaffeNet train_val test mirroring to false
Evan Shelhamer [Tue, 13 Oct 2015 18:09:48 +0000 (11:09 -0700)]
Merge pull request #3187 from akash1810/patch-1
[example] Add pyyaml as a requirement for web demo
Akash A [Tue, 13 Oct 2015 16:53:35 +0000 (17:53 +0100)]
Add pyyaml as a requirement
In getting the [web demo](http://caffe.berkeleyvision.org/gathered/examples/web_demo.html) started I get an `ImportError: No module named yaml` error when running `./scripts/download_model_binary.py models/bvlc_reference_caffenet`.
Ronghang Hu [Tue, 13 Oct 2015 16:36:55 +0000 (09:36 -0700)]
Merge pull request #3184 from lalelale/patch-1
Update store2hdf5.m
Alessandro Giusti [Tue, 13 Oct 2015 12:30:45 +0000 (14:30 +0200)]
Update store2hdf5.m
Fixed a bug in two assertions (the condition input argument must be a scalar logical)
Brian Chu [Tue, 13 Oct 2015 10:50:53 +0000 (03:50 -0700)]
In 00-classification example, get correct class label index
Ronghang Hu [Sun, 11 Oct 2015 16:48:09 +0000 (09:48 -0700)]
Merge pull request #3162 from kkhoot/fix_mvn
Improve numerical stability of variance computation in MVNLayer
Kang Kim [Wed, 7 Oct 2015 16:26:25 +0000 (01:26 +0900)]
Improve numerical stability of variance computation in MVNLayer
Jeff Donahue [Sun, 11 Oct 2015 02:39:22 +0000 (19:39 -0700)]
Merge pull request #3170 from sh1r0/pycaffe_io
Remove the 4D constraint of blobproto IO in python
sh1r0 [Thu, 8 Oct 2015 16:31:05 +0000 (00:31 +0800)]
Remove the 4D constraint of blobproto IO in python
Carl Doersch [Mon, 24 Aug 2015 03:47:25 +0000 (20:47 -0700)]
BatchReindexLayer to shuffle, subsample, and replicate examples in a batch
Ronghang Hu [Wed, 7 Oct 2015 19:41:37 +0000 (12:41 -0700)]
Merge pull request #3164 from eerwitt/layer-documentation-sample-fixes
Documentation References ImageNet Sample Location
e3 [Wed, 7 Oct 2015 18:52:45 +0000 (11:52 -0700)]
fixes BVLC/caffe#3163
Gustav Larsson [Tue, 6 Oct 2015 02:55:00 +0000 (21:55 -0500)]
Add pycaffe test for solver.snapshot()
Jeff Donahue [Mon, 5 Oct 2015 21:36:11 +0000 (14:36 -0700)]
Merge pull request #3152 from jeffdonahue/silence-backward-fix
SilenceLayer Backward bugfix (fixes #3151)
Jeff Donahue [Mon, 5 Oct 2015 21:15:08 +0000 (14:15 -0700)]
SilenceLayer Backward bugfix (fixes #3151)
Jan Issac [Mon, 5 Oct 2015 09:13:26 +0000 (11:13 +0200)]
minor typo fix
Ronghang Hu [Mon, 5 Oct 2015 07:30:10 +0000 (00:30 -0700)]
Merge pull request #3058 from zoharby/master
Add a caffe.io.write_mean function to the MATLAB interface
zoharby [Fri, 11 Sep 2015 12:06:28 +0000 (15:06 +0300)]
Add a caffe.io.write_mean function to the MATLAB interface
Useful for exporting models from MATLAB (e.g. MatConvNet) to Caffe
Evan Shelhamer [Thu, 1 Oct 2015 17:05:32 +0000 (10:05 -0700)]
Merge pull request #3133 from kashefy/badge
add badge for travis build and license
Youssef Kashef [Thu, 1 Oct 2015 16:20:23 +0000 (18:20 +0200)]
add badge for travis build and license
Evan Shelhamer [Thu, 1 Oct 2015 00:43:05 +0000 (17:43 -0700)]
Merge pull request #3069 from timmeinhardt/argmax
Add argmax_param "axis" to maximise output along the specified axis
Ronghang Hu [Tue, 29 Sep 2015 16:07:44 +0000 (09:07 -0700)]
Merge pull request #3128 from hjss06/patch-1
Fix a typo
Yang Song [Tue, 29 Sep 2015 12:07:52 +0000 (20:07 +0800)]
Fix a typo
Fix a typo in the message.
Ronghang Hu [Sat, 26 Sep 2015 18:14:30 +0000 (11:14 -0700)]
Merge pull request #3073 from ronghanghu/consistent-malloc-free
Ensure consistency between memory alloc and free
Luke Yeager [Fri, 25 Sep 2015 22:53:54 +0000 (15:53 -0700)]
Add ALLOW_LMDB_NOLOCK build option
This option lets you open LMDB files with the MDB_NOLOCK flag. You
should not set this flag if you will be reading LMDBs with any
possibility of simultaneous read and write.
Luke Yeager [Fri, 25 Sep 2015 22:43:47 +0000 (15:43 -0700)]
Re-ordering some lines in build files
Enforcing a consistent ordering - OpenCV, LevelDB, LMDB
This will allow me to add the ALLOW_LMDB_NOLOCK option just after the
USE_LMDB option, while keeping the IO dependency options together.
Ronghang Hu [Wed, 16 Sep 2015 20:39:23 +0000 (13:39 -0700)]
Add flag on how host memory is allocated
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.
Evan Shelhamer [Fri, 25 Sep 2015 21:57:17 +0000 (14:57 -0700)]
Merge pull request #3032 from ronghanghu/ban-pythonlayer-in-parallel
Disallow PythonLayer in Multi-GPU training