Kai Li [Tue, 18 Mar 2014 07:22:14 +0000 (15:22 +0800)]
Fixed CPPLint errors related to math funtions
Kai Li [Tue, 11 Mar 2014 09:05:27 +0000 (17:05 +0800)]
Rename signbit in macros to sgnbit to avoid conflicts with std::signbit
Kai Li [Wed, 26 Feb 2014 03:23:20 +0000 (11:23 +0800)]
Add signbit math func, simplify GPU defs & instantiations with a macro
Kai Li [Tue, 25 Feb 2014 12:26:55 +0000 (20:26 +0800)]
Add and test non-in-place scale math functions for CPU and GPU
Kai Li [Tue, 25 Feb 2014 12:06:05 +0000 (20:06 +0800)]
Use macro to simplify element wise cpu math functions
Kai Li [Tue, 25 Feb 2014 11:33:10 +0000 (19:33 +0800)]
Add and test element wise abs math functions for CPU and GPU
Kai Li [Tue, 25 Feb 2014 11:21:18 +0000 (19:21 +0800)]
Instantiate caffe_cpu_sign for float and double
Kai Li [Tue, 25 Feb 2014 11:16:44 +0000 (19:16 +0800)]
Add and test element wise sign math funtions for CPU and GPU
Kai Li [Tue, 25 Feb 2014 10:41:45 +0000 (18:41 +0800)]
Add and test sum of absolute values math functions for CPU and GPU
Evan Shelhamer [Sun, 23 Mar 2014 05:53:42 +0000 (22:53 -0700)]
Merge pull request #165 from BVLC/boost-eigen
MKL/non-MKL Reconciliation
Caffe no longer requires MKL. By default it builds without it, relying on atlas and cblas instead. Set the `USE_MKL` var in your Makefile.config accordingly.
Evan Shelhamer [Sat, 22 Mar 2014 08:27:42 +0000 (01:27 -0700)]
Set copyright to BVLC and contributors.
The exact details of the contributions are recorded by versioning.
Evan Shelhamer [Sat, 22 Mar 2014 06:47:01 +0000 (23:47 -0700)]
Hide boost rng behind facade for osx compatibility
Split boost random number generation from the common Caffe singleton and
add a helper function for rng. This resolves a build conflict in OSX
between boost rng and nvcc compilation of cuda code.
Refer to #165 for a full discussion.
Thanks to @satol for suggesting a random number generation facade rather
than a total split of cpp and cu code, which is far more involved.
Evan Shelhamer [Fri, 21 Mar 2014 22:50:43 +0000 (15:50 -0700)]
lint
Evan Shelhamer [Fri, 21 Mar 2014 21:58:11 +0000 (14:58 -0700)]
clean up residual mkl comments and code
The FIXMEs about RNG were addressed by caffe_nextafter for
uniform distributions and the normal distribution concern is surely a
typo in the boost documentation, since the normal pdf is correctly
stated elsewhere in the documentation.
James Thewlis [Mon, 3 Mar 2014 17:43:20 +0000 (17:43 +0000)]
Added extern C wrapper to cblas.h include
This ensures that it works with ATLAS's header file, which doesn't include such
a guard itself (whereas the reference version from Ubuntu's libblas-dev does)
jamt9000 [Mon, 3 Mar 2014 17:07:23 +0000 (17:07 +0000)]
Fixed order of cblas and atlas linker flags
They were the wrong way round, causing linking to fail in some cases
Evan Shelhamer [Thu, 27 Feb 2014 06:41:58 +0000 (22:41 -0800)]
comment out stray mkl includes
Evan Shelhamer [Tue, 18 Feb 2014 19:10:23 +0000 (11:10 -0800)]
make MKL switch surprise-proof
Evan Shelhamer [Sat, 15 Feb 2014 05:27:20 +0000 (21:27 -0800)]
rewrite MKL flag note, polish makefile
add MKL dirs conditioned on USE_MKL
include libraries before making LD_FLAGS
Rowland Depp [Wed, 12 Feb 2014 05:41:01 +0000 (21:41 -0800)]
major refactoring allow coexistence of MKL and non-MKL cases
Kai Li [Fri, 7 Feb 2014 10:44:10 +0000 (18:44 +0800)]
Replace atlas with multithreaded OpenBLAS to speed-up on multi-core CPU
issue: #79
Jeff Donahue [Wed, 29 Jan 2014 21:11:34 +0000 (13:11 -0800)]
fix bernoulli generator bug
Jeff Donahue [Wed, 29 Jan 2014 21:03:42 +0000 (13:03 -0800)]
add bernoulli rng test to demonstrate bug (generates all 0s unless p ==
1)
Jeff Donahue [Wed, 22 Jan 2014 20:42:12 +0000 (12:42 -0800)]
change all Rng's to use variate_generator for consistency
Jeff Donahue [Wed, 22 Jan 2014 20:28:01 +0000 (12:28 -0800)]
use boost variate_generator to pass tests w/ boost 1.46 (Gaussian filler
previously filled in all NaNs for me, making many tests fail)
Jeff Donahue [Wed, 22 Jan 2014 20:14:09 +0000 (12:14 -0800)]
make uniform distribution usage compatible with boost 1.46
Alejandro Dubrovsky [Wed, 22 Jan 2014 11:56:57 +0000 (22:56 +1100)]
mean_bound and sample_mean need referencing with this
Alejandro Dubrovsky [Wed, 22 Jan 2014 11:56:17 +0000 (22:56 +1100)]
nextafter templates off one type
Evan Shelhamer [Thu, 9 Jan 2014 00:36:52 +0000 (16:36 -0800)]
relax precision of MultinomialLogisticLossLayer test
Kai Li [Sun, 12 Jan 2014 05:55:26 +0000 (13:55 +0800)]
Fix math funcs, add tests, change Eigen Map to unaligned for lrn_layer
[shelhamer: removed math function tests, since they were merged via
other branches]
Kai Li [Sat, 11 Jan 2014 16:39:45 +0000 (00:39 +0800)]
Fix test stochastic pooling stepsize/threshold to be same as max pooling
Kai Li [Sat, 11 Jan 2014 15:57:37 +0000 (23:57 +0800)]
Fixed FlattenLayer Backward_cpu/gpu have no return value
Kai Li [Sat, 11 Jan 2014 15:51:54 +0000 (23:51 +0800)]
Fixed uniform distribution upper bound to be inclusive
Rodrigo Benenson [Sun, 8 Dec 2013 04:55:39 +0000 (15:55 +1100)]
compile caffe without MKL (dependency replaced by boost::random, Eigen3)
- examples, test and pycaffe compile without problem (matcaffe not tested)
- tests show some errors (on cpu gradient tests), to be investigated
- random generators need to be double checked
- mkl commented code needs to be removed
Jeff Donahue [Fri, 21 Mar 2014 20:11:36 +0000 (13:11 -0700)]
Merge pull request #247 from jeffdonahue/loss-in-forward-window-data-layer
Loss in forward pass fix for window data layer
Jeff Donahue [Fri, 21 Mar 2014 20:05:59 +0000 (13:05 -0700)]
loss in forward pass fix for window data layer
Jeff Donahue [Fri, 21 Mar 2014 19:52:16 +0000 (12:52 -0700)]
Merge pull request #209 from jeffdonahue/loss-in-forward-pass
Compute loss in the forward pass
Evan Shelhamer [Thu, 20 Mar 2014 05:34:01 +0000 (22:34 -0700)]
Back-merge documentation and script fixes
fix script path incantation
convert css indentation to spaces
fix cifar10 leveldb creation path
wget without checking certificate for dropbox (dodge complaint on linux)
docs: added list of contributors
minor style update of docs
Evan Shelhamer [Thu, 20 Mar 2014 05:29:17 +0000 (22:29 -0700)]
fix script path incantation
Sergey Karayev [Thu, 20 Mar 2014 05:16:03 +0000 (22:16 -0700)]
convert css indentation to spaces
Evan Shelhamer [Thu, 20 Mar 2014 05:13:24 +0000 (22:13 -0700)]
fix cifar10 leveldb creation path
Evan Shelhamer [Thu, 20 Mar 2014 05:01:45 +0000 (22:01 -0700)]
wget without checking certificate for dropbox (dodge complaint on linux)
Sergey Karayev [Thu, 20 Mar 2014 04:39:52 +0000 (21:39 -0700)]
docs: added list of contributors
Sergey Karayev [Thu, 20 Mar 2014 04:27:22 +0000 (21:27 -0700)]
minor style update of docs
Sergey Karayev [Thu, 20 Mar 2014 04:25:44 +0000 (21:25 -0700)]
Fix to #161
- signficantly change the documentation file
- link to it from index.md
- remove the image resizing script, since (a) it does not work, (b) is obviated by using ImagesLayer
- add sample prototxt that uses ImagesLayer.
Sergey Karayev [Thu, 20 Mar 2014 04:21:28 +0000 (21:21 -0700)]
Merge pull request #161 from kloudkl/simplify_feature_extraction
Feature extraction, feature binarization and image retrieval examples
Jeff Donahue [Thu, 20 Mar 2014 03:50:13 +0000 (20:50 -0700)]
minor cleanup in rcnn-finetuning -- rcnn feature computation tested at
this commit (in addition to all caffe unit tests passing)
Ross Girshick [Tue, 4 Feb 2014 19:00:55 +0000 (11:00 -0800)]
cleanup matlab demo
Ross Girshick [Mon, 13 Jan 2014 21:34:34 +0000 (13:34 -0800)]
add initialization key for verifying state
Ross Girshick [Fri, 20 Dec 2013 22:39:22 +0000 (14:39 -0800)]
demo on how to get net weights using the matlab interface
Ross Girshick [Fri, 6 Dec 2013 04:58:03 +0000 (20:58 -0800)]
return model weights
Evan Shelhamer [Thu, 13 Mar 2014 22:19:43 +0000 (15:19 -0700)]
keep DLOG (revert accidental switch to LOG)
Evan Shelhamer [Thu, 13 Mar 2014 22:10:41 +0000 (15:10 -0700)]
file pascal finetuning prototxt examples and fix paths
Ross Girshick [Fri, 24 Jan 2014 00:56:07 +0000 (16:56 -0800)]
set default to the best value
Ross Girshick [Wed, 22 Jan 2014 07:23:33 +0000 (23:23 -0800)]
some cleanup
Ross Girshick [Fri, 17 Jan 2014 00:25:12 +0000 (16:25 -0800)]
fix paths
Ross Girshick [Fri, 17 Jan 2014 00:22:08 +0000 (16:22 -0800)]
support for tightest square mode while finetuning
Ross Girshick [Sun, 12 Jan 2014 15:54:35 +0000 (07:54 -0800)]
10x learning rate for fine tuning makes a big difference
Ross Girshick [Thu, 9 Jan 2014 21:48:50 +0000 (13:48 -0800)]
support for adding padding to windows in the window_data_layer
Ross Girshick [Tue, 7 Jan 2014 23:04:19 +0000 (15:04 -0800)]
Code that was used to finetune with reasonable success
Ross Girshick [Mon, 23 Dec 2013 19:52:03 +0000 (11:52 -0800)]
some major bug fixes (includes some to-be-removed debugging code)
Ross Girshick [Fri, 20 Dec 2013 22:32:55 +0000 (14:32 -0800)]
adjustments to try to match the setup for fine tuning with cuda-convnet
Ross Girshick [Thu, 13 Mar 2014 23:35:27 +0000 (16:35 -0700)]
define pascal finetuning models
Ross Girshick [Thu, 13 Mar 2014 23:35:10 +0000 (16:35 -0700)]
add window data layer
Jeff Donahue [Wed, 19 Mar 2014 01:55:56 +0000 (18:55 -0700)]
post rebase fixes: images layer and padding layer compute loss in
forward
Jeff Donahue [Sat, 15 Mar 2014 19:22:53 +0000 (12:22 -0700)]
null pointer defaults for forward loss outputs
Jeff Donahue [Fri, 14 Mar 2014 21:52:51 +0000 (14:52 -0700)]
loss in forward pass for concat layer (thought i'd rebased to latest dev
but apparently not)
Jeff Donahue [Fri, 14 Mar 2014 05:44:41 +0000 (22:44 -0700)]
fix softmax loss layer bug; all tests pass
Jeff Donahue [Fri, 14 Mar 2014 01:27:46 +0000 (18:27 -0700)]
remove accidentally added empty line
Jeff Donahue [Fri, 14 Mar 2014 01:16:21 +0000 (18:16 -0700)]
revert unnecessary reordering of lines in softmaxwithlosslayer backward
Jeff Donahue [Fri, 14 Mar 2014 01:09:52 +0000 (18:09 -0700)]
gradient checker optimization with forward pass loss: only need to run
backward pass to compute analytic gradient (the thing being checked) now
Jeff Donahue [Fri, 14 Mar 2014 01:08:34 +0000 (18:08 -0700)]
test_gradient_check_util: blobid -> blob_id
Jeff Donahue [Fri, 14 Mar 2014 01:01:15 +0000 (18:01 -0700)]
make tests compile and pass
Jeff Donahue [Fri, 14 Mar 2014 00:54:10 +0000 (17:54 -0700)]
fix net_speed_benchmark so 'make all' works
Jeff Donahue [Fri, 14 Mar 2014 00:50:15 +0000 (17:50 -0700)]
change specification of forward/backward function and fix layer
definitions appropriately
Evan Shelhamer [Wed, 19 Mar 2014 16:56:20 +0000 (09:56 -0700)]
Merge pull request #221 from jamt9000/fix-dump-network
Fix dump network
Evan Shelhamer [Wed, 19 Mar 2014 16:36:55 +0000 (09:36 -0700)]
Merge pull request #242 from kloudkl/script
Fix shell scripts: current dir, wget output, and naming conventions
Kai Li [Wed, 19 Mar 2014 15:26:14 +0000 (23:26 +0800)]
File naming convention requires that two words be split by a underscore
Kai Li [Wed, 19 Mar 2014 15:21:19 +0000 (23:21 +0800)]
Wget should never be quiet
Kai Li [Wed, 19 Mar 2014 15:20:09 +0000 (23:20 +0800)]
Fix shell script current dir bad substition errors
Kai Li [Wed, 19 Mar 2014 15:57:39 +0000 (23:57 +0800)]
Explain how to get the mean image of ILSVRC
Kai Li [Wed, 19 Mar 2014 15:52:26 +0000 (23:52 +0800)]
Change generate file list python script path in feature extraction doc
Kai Li [Wed, 19 Mar 2014 15:08:51 +0000 (23:08 +0800)]
Removing feature binarization and image retrieval examples
Kai Li [Tue, 18 Mar 2014 07:40:23 +0000 (15:40 +0800)]
Move binarize_features, retrieve_images to examples/feauture_extraction
Kai Li [Mon, 17 Mar 2014 10:09:57 +0000 (18:09 +0800)]
Add documentation for the feature extraction demo
Kai Li [Mon, 17 Mar 2014 09:37:02 +0000 (17:37 +0800)]
Add a python script to generate a list of all the files in a directory
Kai Li [Mon, 17 Mar 2014 08:14:01 +0000 (16:14 +0800)]
Don't create a new batch after all the feature vectors have been saved
Kai Li [Wed, 26 Feb 2014 14:07:48 +0000 (22:07 +0800)]
Fix cpplint errors for Net, its tests and feature related 3 examples
Kai Li [Wed, 26 Feb 2014 13:37:25 +0000 (21:37 +0800)]
Use lowercase underscore naming convention for Net blob & layer getters
Kai Li [Wed, 26 Feb 2014 04:29:37 +0000 (12:29 +0800)]
Move extract_features, binarize_features, retrieve_images to tools/
Kai Li [Wed, 26 Feb 2014 01:30:35 +0000 (09:30 +0800)]
Save and load data correctly in feat extracion, binarization and IR demo
Kai Li [Tue, 25 Feb 2014 22:51:32 +0000 (06:51 +0800)]
Change feature binarization threshold to be the mean of all the values
rather than zero in the feature binarization example
Kai Li [Tue, 25 Feb 2014 21:55:58 +0000 (05:55 +0800)]
Fix saving real valued feature bug in the feature extraction example
Kai Li [Tue, 25 Feb 2014 21:34:23 +0000 (05:34 +0800)]
Fix bugs in the image retrieval example
Kai Li [Tue, 25 Feb 2014 19:47:32 +0000 (03:47 +0800)]
Fix bugs of the feature binarization example
Kai Li [Tue, 25 Feb 2014 19:46:32 +0000 (03:46 +0800)]
Enhance help, log message & format of the feature extraction example
Kai Li [Tue, 25 Feb 2014 18:47:08 +0000 (02:47 +0800)]
Fix bugs in the feature extraction example
Kai Li [Tue, 25 Feb 2014 10:23:04 +0000 (18:23 +0800)]
Add __builtin_popcount* based fast Hamming distance math function
Kai Li [Sun, 23 Feb 2014 14:26:35 +0000 (22:26 +0800)]
Simplify image retrieval example to use binary features directly
Kai Li [Sun, 23 Feb 2014 14:25:58 +0000 (22:25 +0800)]
Add feature binarization example