SergioRAgostinho
commented
8 years ago So, I just hit the same problem while invoking
caffe.run_tests()in Matlab. The "offending" line is this one. The content of the file passed in solver_file is similar to this.
net: "/tmp/tp5f07d5a6_88ae_4365_b574_d20a69a66765"
test_iter: 10 test_interval: 10 base_lr: 0.01 momentum: 0.9
weight_decay: 0.0005 lr_policy: "inv" gamma: 0.0001 power: 0.75
display: 100 max_iter: 100 snapshot_after_train: falseThe stacktrace in matlab is
K>> dbstack
> In caffe.get_solver (line 7)
In caffe.Solver (line 21)
In caffe.test.test_solver (line 22)
In caffe.run_tests (line 12)going through here and eventually dying here. The registry is empty and the thing aborts with
...
I0721 18:32:09.077687 25952 net.cpp:283] Network initialization done.
F0721 18:34:51.517117 25952 solver_factory.hpp:76] Check failed: registry.count(type) == 1 (0 vs. 1) Unknown solver type: SGD (known types: )
*** Check failure stack trace: ***The build options are below.
-- ******************* Caffe Configuration Summary *******************
-- General:
-- Version : 1.0.0-rc3
-- Git : rc3-216-g42cd785-dirty
-- System : Linux
-- C++ compiler : /usr/bin/c++
-- Release CXX flags : -O3 -DNDEBUG -fPIC -Wall -Wno-sign-compare -Wno-uninitialized
-- Debug CXX flags : -g -fPIC -Wall -Wno-sign-compare -Wno-uninitialized
-- Build type : Release
--
-- BUILD_SHARED_LIBS : ON
-- BUILD_python : ON
-- BUILD_matlab : ON
-- BUILD_docs : ON
-- CPU_ONLY : OFF
-- USE_OPENCV : ON
-- USE_LEVELDB : ON
-- USE_LMDB : ON
-- ALLOW_LMDB_NOLOCK : OFF
--
-- Dependencies:
-- BLAS : Yes (Open)
-- Boost : Yes (ver. 1.55)
-- glog : Yes
-- gflags : Yes
-- protobuf : Yes (ver. 2.5.0)
-- lmdb : Yes (ver. 0.9.10)
-- LevelDB : Yes (ver. 1.15)
-- Snappy : Yes (ver. 1.1.0)
-- OpenCV : Yes (ver. 2.4.8)
-- CUDA : Yes (ver. 7.5)
--
-- NVIDIA CUDA:
-- Target GPU(s) : Auto
-- GPU arch(s) : sm_52
-- cuDNN : Yes (ver. 5.0.5)
--
-- Matlab:
-- Matlab : Yes (/opt/MATLAB/R2015a/bin/mex, /opt/MATLAB/R2015a/bin/mexext
-- Octave : No
--
-- Documentaion:
-- Doxygen : /usr/bin/doxygen (1.8.6)
-- config_file : /<omitted>/caffe/.Doxyfile
--
-- Install:
-- Install path : /usr/local
--
-- Configuring done
-- Generating done
-- Build files have been written to: /<omitted>/caffe/buildI was trying to see if I could patch this but this registry mechanism is somewhat cryptic at first sight. Any quick pointers on how and when do the solvers get registered?
ghostcow
rokm
willyd
rcorbish
spark123456789
rajasrijan
soulslicer
lizhongguo
This is a similar problem to #3712 - namely matcaffe complaining at runtime that it doesn't know any concrete Solver types.
Caffe builds successfully with CUDA and the matlab wrapper, but when I try to call caffe.get_solver(some_prototxt_file) from Matlab, I'm getting an exception thrown with the error text:
F0417 22:03:48.503923 11578 solver_factory.hpp:76] Check failed: registry.count(type) == 1 (0 vs. 1) Unknown solver type: SGD (known types: )
This is obviously related to the issues raised in:
https://groups.google.com/forum/#!topic/caffe-users/TMuS7A_mccI http://stackoverflow.com/questions/32416226/create-layer-error-at-layer-factory-hpp https://groups.google.com/forum/#!topic/caffe-users/Py6IwMQvtqo/discussion http://stackoverflow.com/questions/30325108/caffe-layer-creation-failure
where the common factor seems to be the usage of Caffe as a statically linked lib rather than a dynamically linked lib. This is not true in my case...as confirmed by ldd, which shows the second run time dependency of the compiled mex-file to be libcaffe.so.1.0.0-rc3. Furthermore, I don't even have any *.a files in the entire caffe project folder.
Also, I can't see a SolverFactory or anything like LayerFactory, where different types of layer would need registering as in http://stackoverflow.com/questions/32416226/create-layer-error-at-layer-factory-hpp
I call cmake in relative path ./build with:
cmake .. -DCMAKE_INSTALL_PREFIX=$MINICONDA_ENV -DCUDA_TOOLKIT_ROOT_DIR=/vol/cuda/7.5.18 -DUSE_LMDB=0 -DUSE_CUDNN=0 -DUSE_LEVELDB=0 -DCPU_ONLY=0 -DBUILD_python=0 -DBUILD_docs=0 -DBLAS=atlas -DBUILD_matlab=1 && make -j6
This is the only thing I'm doing different from the installation tutorial (rather than using the Makefile.config)