Evan Shelhamer [Fri, 16 Jan 2015 05:04:43 +0000 (21:04 -0800)]
Merge pull request #1236 from mlapin/legacy_nvcc_support
Drop OpenCV includes from NVCC code for legacy reasons.
Evan Shelhamer [Fri, 16 Jan 2015 04:49:50 +0000 (20:49 -0800)]
Merge pull request #1740 from shelhamer/yosemite-makefile
Support OS X Yosemite / 10.10
Evan Shelhamer [Fri, 16 Jan 2015 00:28:09 +0000 (16:28 -0800)]
support OS X Yosemite / 10.10
- pick libstdc++ for OS X (regardless of version)
- make gtest rely on its own tuple to not conflict with clang
(thanks @pluskid!)
- 10.10 has Accelerate while 10.9 has vecLib for BLAS
(thanks @leonardt and @drdan14)
Jonathan L Long [Sat, 10 Jan 2015 09:16:12 +0000 (01:16 -0800)]
fix typo in layer_factory.cpp
Jeff Donahue [Wed, 7 Jan 2015 01:04:49 +0000 (17:04 -0800)]
Merge pull request #1228 from longjon/solver-step
Refactor Solver to allow interactive stepping
Jonathan L Long [Mon, 29 Sep 2014 23:49:49 +0000 (16:49 -0700)]
[pycaffe] expose SGDSolver.step
Jonathan L Long [Mon, 29 Sep 2014 23:47:06 +0000 (16:47 -0700)]
break out Step from Solver
Jeff Donahue [Wed, 31 Dec 2014 00:02:31 +0000 (16:02 -0800)]
Merge pull request #1661 from longjon/init-test-labels
Fix sketchy label initialization in NetTest
Jonathan L Long [Tue, 30 Dec 2014 20:52:53 +0000 (12:52 -0800)]
[tests] don't use Gaussian labels in NetTest's TinyNet
Previously, labels were being nonsensically filled with a zero-mean
Gaussian. Now they are just set to constant zero. Note that this is not
necessary for TrickyNet, neither forward nor backward are ever called.
Jon Long [Tue, 30 Dec 2014 10:33:43 +0000 (02:33 -0800)]
Merge pull request #1658 from jeffdonahue/make-everything-tweaks
'make everything' tweaks
Jeff Donahue [Tue, 30 Dec 2014 09:33:05 +0000 (01:33 -0800)]
don't do runtest as part of 'make everything'
Jeff Donahue [Tue, 30 Dec 2014 09:31:46 +0000 (01:31 -0800)]
only build matcaffe as part of 'make everything' if MATLAB_DIR is set
Jon Long [Tue, 30 Dec 2014 04:49:14 +0000 (20:49 -0800)]
Merge pull request #1656 from longjon/fix-softmax-loss-stubs
Fix an error made in removing the SoftmaxLossLayer GPU stubs
Jonathan L Long [Tue, 30 Dec 2014 01:59:07 +0000 (17:59 -0800)]
remove SoftmaxLossLayer CPU_ONLY stubs, since there is no GPU version
Evan Shelhamer [Tue, 30 Dec 2014 01:48:07 +0000 (20:48 -0500)]
Merge pull request #1655 from longjon/softmax-loss-todo
Move softmax loss GPU todo comment from .cu file to header file
Jonathan L Long [Tue, 30 Dec 2014 01:36:08 +0000 (17:36 -0800)]
move softmax loss GPU todo comment to header file
There's no need to have GPU stubs just for a comment, and the absence of
a GPU-specific implementation is made more obvious by the nonexistence
of the .cu file.
Jon Long [Mon, 29 Dec 2014 22:57:41 +0000 (14:57 -0800)]
Merge pull request #1648 from longjon/find-warnings
Fix find warnings in upload_model_to_gist.sh
Jon Long [Mon, 29 Dec 2014 05:39:27 +0000 (21:39 -0800)]
Merge pull request #1561 from longjon/pretty-build
Pretty up the build
Jonathan L Long [Mon, 29 Dec 2014 04:57:37 +0000 (20:57 -0800)]
[scripts] fix find warnings in upload_model_to_gist.sh
Jeff Donahue [Mon, 29 Dec 2014 02:36:19 +0000 (18:36 -0800)]
Merge pull request #1645 from longjon/remove-get-layer
Remove unused GetLayer function (replaced by LayerRegistry::CreateLayer)
Jonathan L Long [Mon, 29 Dec 2014 02:03:45 +0000 (18:03 -0800)]
remove unused GetLayer function (replaced by LayerRegistry::CreateLayer)
Jonathan L Long [Thu, 11 Dec 2014 22:45:38 +0000 (14:45 -0800)]
pretty the build with the Q variable
Unset Q (in the environment) to see full commands, or edit in
Makefile.config.
Jon Long [Mon, 29 Dec 2014 01:36:24 +0000 (17:36 -0800)]
Merge pull request #1472 from longjon/incremental-build
Makefile improvements: rule consolidation, dependency generation
Jonathan L Long [Mon, 24 Nov 2014 21:37:06 +0000 (13:37 -0800)]
automatic dependency generation
Yangqing Jia [Sun, 28 Dec 2014 21:34:43 +0000 (13:34 -0800)]
Merge pull request #1636 from longjon/get-layer-gone
Holiday merge - Update use of GetLayer to new registry function
Jonathan L Long [Sat, 27 Dec 2014 09:47:59 +0000 (01:47 -0800)]
update use of GetLayer -> LayerRegistry::CreateLayer
Sergio Guadarrama [Fri, 26 Dec 2014 19:13:40 +0000 (11:13 -0800)]
Merge pull request #1585 from longjon/check-malloc
Check host malloc result
Evan Shelhamer [Sun, 21 Dec 2014 08:43:26 +0000 (00:43 -0800)]
Merge pull request #1608 from sguada/added_credits
Add credits for training BVLC Models
Sergio [Sun, 21 Dec 2014 07:28:19 +0000 (23:28 -0800)]
Added credits and bvlc_googlenet to model_zoo.md
Sergio [Sun, 21 Dec 2014 07:24:32 +0000 (23:24 -0800)]
Added credits for training bvlc models
Sergio Guadarrama [Sun, 21 Dec 2014 05:13:40 +0000 (21:13 -0800)]
Merge pull request #1598 from sguada/bvlc_googlenet
Added bvlc_googlenet prototxt and weights
Sergio [Tue, 2 Dec 2014 21:10:57 +0000 (13:10 -0800)]
Added bvlc_googlenet prototxt and weights
Jonathan L Long [Wed, 17 Dec 2014 04:18:53 +0000 (20:18 -0800)]
check host malloc result
Sergio Guadarrama [Tue, 16 Dec 2014 23:32:42 +0000 (15:32 -0800)]
Merge pull request #1558 from hojonathanho/hdf5_error_msg
Check input line count in HDF5 data layer
Jonathan L Long [Thu, 11 Dec 2014 07:37:42 +0000 (23:37 -0800)]
move cuda output from build/.../.cuo -> build/cuda/.../.o
This will allow nvcc's -M dependency generation option to work
harmoniously, since it assumes that output will have a .o extension.
Jonathan L Long [Mon, 24 Nov 2014 08:10:35 +0000 (00:10 -0800)]
automatic discovery of source directories
Jonathan L Long [Mon, 24 Nov 2014 06:35:46 +0000 (22:35 -0800)]
consolidate build rules for object files
Jonathan L Long [Tue, 16 Dec 2014 09:52:43 +0000 (01:52 -0800)]
remove extra blank line
Jonathan Ho [Thu, 11 Dec 2014 02:27:56 +0000 (18:27 -0800)]
Check input line count in HDF5 data layer
Evan Shelhamer [Mon, 8 Dec 2014 22:00:49 +0000 (17:00 -0500)]
Merge pull request #1384 from CellScope/log-parser-python
Python log parser
Daniel Golden [Mon, 3 Nov 2014 05:02:07 +0000 (21:02 -0800)]
Store data in lists of dicts and use csv package
Output format is unchanged (except that csv.DictWriter insists on writing ints as 0.0 instead of 0)
Daniel Golden [Fri, 31 Oct 2014 17:24:07 +0000 (10:24 -0700)]
Take train loss from `Iteration N, loss = X` lines
Was previously using `Train net output #M: loss = X` lines, but there may not be exactly one of those (e.g., for GoogLeNet, which has multiple loss layers); I believe that `Iteration N, loss = X` is the aggregated loss.
If there's only one loss layer, these two values will be equal and it won't matter. Otherwise, we presumably want to report the aggregated loss.
Daniel Golden [Thu, 30 Oct 2014 22:51:14 +0000 (15:51 -0700)]
Created parse_log.py, competitor to parse_log.sh
Evan Shelhamer [Mon, 8 Dec 2014 19:50:02 +0000 (14:50 -0500)]
Merge pull request #1406 from CellScope/matcaffe-osx-fix-crash-on-error
Prevent Matlab on OS X from crashing on error
Evan Shelhamer [Mon, 8 Dec 2014 19:45:50 +0000 (14:45 -0500)]
Merge pull request #1516 from drdan14/update-homebrew-install-instructions
Better instructions for updating Homebrew after modifying formulae
Evan Shelhamer [Mon, 8 Dec 2014 19:44:15 +0000 (14:44 -0500)]
Merge pull request #1527 from drdan14/classify-py-mean-dims
Update mean file help
Jonathan L Long [Sun, 7 Dec 2014 11:06:35 +0000 (03:06 -0800)]
clarify #endif comment
Daniel Golden [Wed, 3 Dec 2014 22:30:14 +0000 (14:30 -0800)]
Update mean file help
Numpy array dimensions should be [Channels x Height x Width], NOT [Height x Width x Channels]
If an image mean file is supplied with channels as the last dimension, classify.py throws errors like:
ValueError: operands could not be broadcast together with shapes (3,227,227) (256,0,225) (3,227,227)
Daniel Golden [Tue, 2 Dec 2014 19:20:58 +0000 (11:20 -0800)]
Better instructions for updating Homebrew after modifying formulae
Previous instructions glossed over how to deal with OpenCV-related update problems
Jeff Donahue [Mon, 1 Dec 2014 19:33:51 +0000 (11:33 -0800)]
Merge pull request #1505 from longjon/conv-dedup
Remove redundant code in ConvolutionLayer::Reshape
Jonathan L Long [Mon, 1 Dec 2014 05:34:29 +0000 (21:34 -0800)]
remove redundant code in ConvolutionLayer::Reshape
Jeff Donahue [Wed, 26 Nov 2014 00:17:57 +0000 (16:17 -0800)]
Merge pull request #1455 from seanbell/tanh-fix
Fix tanh layer for badly scaled input values
Jeff Donahue [Mon, 24 Nov 2014 02:29:24 +0000 (18:29 -0800)]
Merge pull request #1469 from longjon/label-lower-bound
Check both lower and upper bounds in SoftmaxLossLayer
Jonathan L Long [Mon, 24 Nov 2014 01:58:24 +0000 (17:58 -0800)]
use DCHECK in SoftmaxLossLayer so that bounds checking has no perf penalty without DEBUG
Jonathan L Long [Sat, 22 Nov 2014 19:59:07 +0000 (11:59 -0800)]
in SoftmaxLossLayer, check label >= 0 in addition to upper bound
Sean Bell [Thu, 20 Nov 2014 04:21:53 +0000 (23:21 -0500)]
Fixed header order to satisfy linter
Sean Bell [Wed, 19 Nov 2014 22:25:08 +0000 (17:25 -0500)]
fixed tanh to not return NaN for input values outside the range [-40, 40]
Yangqing Jia [Fri, 14 Nov 2014 06:14:38 +0000 (22:14 -0800)]
Merge pull request #1432 from Yangqing/dev
fix relu cudnn test bug
Yangqing Jia [Fri, 14 Nov 2014 06:05:42 +0000 (22:05 -0800)]
fix relu cudnn test bug
Yangqing Jia [Sat, 8 Nov 2014 02:29:34 +0000 (18:29 -0800)]
Merge pull request #1417 from Yangqing/dev
clean incorrect relu test code
Yangqing Jia [Sat, 8 Nov 2014 01:42:43 +0000 (17:42 -0800)]
relax benchmark milliseconds threshold
Yangqing Jia [Sat, 8 Nov 2014 01:17:49 +0000 (17:17 -0800)]
clean incorrect relu test code
Daniel Golden [Wed, 5 Nov 2014 21:36:00 +0000 (13:36 -0800)]
Prevent Matlab on OS X from crashing on error
Replace CHECK() and LOG(FATAL) with LOG(ERROR) and mexErrMsgTxt
A failed CHECK() or LOG(FATAL) causes Matlab to crash on OS X 10.9 with Matlab 2014a.
Yangqing Jia [Thu, 23 Oct 2014 06:28:46 +0000 (23:28 -0700)]
Merge pull request #1344 from baeuml/minor-whitespace-in-logging-message-fix
Minor whitespace fix in logging message in HDF5 output layer
Martin Baeuml [Wed, 22 Oct 2014 08:19:44 +0000 (10:19 +0200)]
Minor whitespace fix in logging message in HDF5 output layer
Evan Shelhamer [Tue, 21 Oct 2014 17:19:56 +0000 (10:19 -0700)]
make release, debug build dirs configurable in Makefile
Evan Shelhamer [Tue, 21 Oct 2014 17:17:11 +0000 (10:17 -0700)]
groom ignore
- re-group general and Caffe ignores
- ignore OS X files and Eclipse .settings as suggested by @cypof
Yangqing Jia [Mon, 20 Oct 2014 22:16:55 +0000 (15:16 -0700)]
Update data_transformer.hpp
Minor grammar fix in comments.
Evan Shelhamer [Mon, 20 Oct 2014 16:25:27 +0000 (09:25 -0700)]
Merge pull request #1326 from jackculpepper/absgradatzero
[fix] define gradient at zero to be zero to fix ABSVAL NaN
Maksim Lapin [Tue, 7 Oct 2014 19:48:26 +0000 (21:48 +0200)]
include opencv only in .cpp
Jack Culpepper [Sun, 19 Oct 2014 15:31:29 +0000 (08:31 -0700)]
define gradient at zero to be zero
Evan Shelhamer [Sun, 19 Oct 2014 02:36:50 +0000 (19:36 -0700)]
Merge pull request #1319 from kmatzen/lmdb_iter_fix
LMDB read-only transaction limit fix
Sergio Guadarrama [Sat, 18 Oct 2014 15:07:31 +0000 (08:07 -0700)]
Merge pull request #1320 from sguada/remove_TIMING
Remove TIMING from ForwardBackward
Sergio [Sat, 18 Oct 2014 14:59:53 +0000 (07:59 -0700)]
Remove TIMING from ForwardBackward
Kevin James Matzen [Sat, 18 Oct 2014 06:51:29 +0000 (02:51 -0400)]
LMDB doesn't support many concurrent read-only transactions, so this preallocates one read-only transaction and reuses it. It's very important that iterations are considered invalid after a commit has been performed.
Sergey Karayev [Fri, 17 Oct 2014 18:51:27 +0000 (11:51 -0700)]
back-merge
Sergey Karayev [Fri, 17 Oct 2014 18:50:48 +0000 (11:50 -0700)]
Merge pull request #1315 from sergeyk/master
Fixing finetune_flickr_style model reported accuracy.
Sergey Karayev [Fri, 17 Oct 2014 18:34:25 +0000 (11:34 -0700)]
Fixing finetune_flickr_style model reported accuracy.
Sergio Guadarrama [Thu, 16 Oct 2014 20:35:17 +0000 (13:35 -0700)]
Merge pull request #1308 from sguada/new_lr_policies
[accidental PR: step size fix]
Evan Shelhamer [Thu, 16 Oct 2014 20:02:01 +0000 (13:02 -0700)]
Merge pull request #1309 from CellScope/edit-brew-boost-python
Edit boost-python formula for OS X 10.9 compatibility
Daniel Golden [Thu, 16 Oct 2014 19:55:53 +0000 (12:55 -0700)]
Edit boost-python formula
Sergio [Thu, 16 Oct 2014 18:17:24 +0000 (11:17 -0700)]
Fix init current_step
Yangqing Jia [Thu, 16 Oct 2014 17:58:37 +0000 (10:58 -0700)]
Merge pull request #1296 from crizCraig/patch-5
Sometimes anaconda is installed in root.
Sergio Guadarrama [Thu, 16 Oct 2014 17:48:23 +0000 (10:48 -0700)]
Merge pull request #1293 from sguada/new_lr_policies
Fixed paths for Added Multistep, Poly and Sigmoid learning rate decay policies
Craig Quiter [Thu, 16 Oct 2014 04:26:01 +0000 (21:26 -0700)]
Sometimes anaconda is installed in root.
Helps fix #489.
Sergio [Sat, 4 Oct 2014 00:14:20 +0000 (17:14 -0700)]
Adapt lenet_multistep_solver.prototxt to current solvers
Sergio Guadarrama [Thu, 16 Oct 2014 01:36:59 +0000 (18:36 -0700)]
Merge pull request #190 from sguada/new_lr_policies
New lr policies, MultiStep and StepEarly
Sergio Guadarrama [Thu, 16 Oct 2014 00:17:05 +0000 (17:17 -0700)]
Merge pull request #1239 from sguada/encoded
Allow using encoded images in Datum, LevelDB, LMDB
Sergio [Mon, 13 Oct 2014 06:16:59 +0000 (23:16 -0700)]
Added CPUTimer
Make timing more precise using double and microseconds
Sergio [Wed, 15 Oct 2014 22:36:42 +0000 (15:36 -0700)]
Upgrade compute_image_mean to use gflags, accept list_of_images, and print mean_values
Sergio [Wed, 15 Oct 2014 22:36:16 +0000 (15:36 -0700)]
Change caffe time to do forward/backward and accumulate time per layer
Sergio [Wed, 8 Oct 2014 00:19:15 +0000 (17:19 -0700)]
Added cache_images to WindowDataLayer
Added root_folder to WindowDataLayer to locate images
Sergio [Tue, 7 Oct 2014 21:14:50 +0000 (14:14 -0700)]
Speed up WindowDataLayer and add mean_values
Sergio [Tue, 7 Oct 2014 18:19:16 +0000 (11:19 -0700)]
Add root_folder to ImageDataLayer
Sergio [Wed, 15 Oct 2014 22:35:26 +0000 (15:35 -0700)]
Add fast code for transform(cv::Mat,Blob)
Sergio [Wed, 15 Oct 2014 22:35:01 +0000 (15:35 -0700)]
Added timers for benchmarking
Sergio [Tue, 7 Oct 2014 06:27:40 +0000 (23:27 -0700)]
Added test for encoded Datum to test_io.cpp
Sergio [Tue, 7 Oct 2014 01:25:28 +0000 (18:25 -0700)]
Added encoded datum to io
Sergio [Tue, 7 Oct 2014 01:05:11 +0000 (18:05 -0700)]
Added encoded option and check_size to convert_imageset
Conflicts:
tools/convert_imageset.cpp
Sergio Guadarrama [Wed, 15 Oct 2014 22:54:11 +0000 (15:54 -0700)]
Merge pull request #1288 from sguada/first_last
Added first_key and last_key to dataset