superres

It's currently very complicated to tie FrameSource to SuperResolution in Lua. Probably we should change PtrWrapper to point to Ptr<SomeAlgorithm> rather than to SomeAlgorithm.

Multichannel CUDA Mats fail to convert

Creating this as a reminder to myself
See demo/cuda/resize.lua (now works only with grayscale images)

imgproc tests

Priority testing:

undistort (how should it work? please test if it's correct)
grabCut (same thing)
Subdiv2D:getTriangleList
Subdiv2D:getEdgeList
Subdiv2D:getVoronoiFacetList

Class-based stuff including algorithms:

Separate functions (filters, detectors, ...):

imgproc

Class-based stuff including algorithms:

Separate functions (filters, detectors, ...):

optflow

I couldn't find corresponding OpenCV include files (candidates are superres/optical_flow.hpp, videostab/optical_flow.hpp and video/tracking.hpp). Current docs on optflow don't match with these!

Scanning dependencies of target optflow
[ 65%] Building CXX object CMakeFiles/optflow.dir/src/optflow.cpp.o
In file included from /Users/eugenioculurciello/Code/Lua/torch-opencv/src/optflow.cpp:1:
In file included from /Users/eugenioculurciello/Code/Lua/torch-opencv/include/optflow.hpp:3:
/Users/eugenioculurciello/Code/Lua/torch-opencv/include/video.hpp:4:10: fatal error: 'opencv2/bgsegm.hpp' file not found

include <opencv2/bgsegm.hpp>

1 error generated.
make[2]: *** [CMakeFiles/optflow.dir/src/optflow.cpp.o] Error 1
make[1]: *** [CMakeFiles/optflow.dir/all] Error 2
make: *** [all] Error 2

video tests

cudaoptflow

Setting a custom optical flow algorithm for SuperResolution

...results in a segfault when calling superRes->nextFrame(). Otherwise it sets Farneback optical flow by default and works fine.

I'm stuck deep in debug with that, couldn't figure out the root of the problem. To be fixed later, possibly by someone else.

build fails: "SSE4.1 instruction set not enabled"

I tried to build torch-opencv on Ubuntu 14.04.3 LTS:

In file included from /home/koepf/torch/install/include/TH/THVector.h:24:0,
                 from /home/koepf/torch/install/include/TH/TH.h:12,
                 from /home/koepf/code/torch-opencv/include/Common.hpp:5,
                 from /home/koepf/code/torch-opencv/src/Common.cpp:1:
/usr/lib/gcc/x86_64-linux-gnu/4.8/include/smmintrin.h:31:3: error: #error "SSE4.1 instruction set not enabled"
 # error "SSE4.1 instruction set not enabled"

Any ideas?

photo

Tonemap

constructor
process
getGamma
setGamma

TonemapDrago

TonemapDurand

TonemapReinhard

TonemapMantiuk

AlignExposures

process

AlignMTB

CalibrateCRF

process

CalibrateDebevec

CalibrateRobertson

MergeExposures

process

MergeDebevec

constructor
process

MergeMertens

MergeRobertson

constructor
process

flann

cudaimgproc

input tensor image into cv

Hi, if I have an image already in a tensor, how can I convert it to opencv format?

flann tests

To-do: copy to-do list from #19

video

BackgroundSubtractor

apply
getBackgroundImage

BackgroundSubtractorMOG2

BackgroundSubtractorKNN

cudacodec

Add rockspec file

working on it myself

demo/videocap.lua issue with cv.cvtcolor

Hello,

I tried to execute the demo/videocap.lua using opencv 3.0 as specified and it's not working properly.

I found the problem to be in line 20:
edges = cv.cvtColor{frame, cv.COLOR_BGR2GRAY}

the problem is that "edges" needs to passed as an argument also:
edges = cv.cvtColor{frame, edges, cv.COLOR_BGR2GRAY}

and now everything works fine.
Just posted the issue, in case anyone else encounters the same problem.

Thank you!

cudaarithm

Installation problem: Optflow

I am encountering the following error probably because some functions were moved into opencv_contrib. I am currently using the command:
OpenCV_DIR="/home/karan/opencv" luarocks install cv

How do I include contrib folder?

Error that I got:

/tmp/luarocks_cv-scm-1-490/torch-opencv/src/optflow.cpp: In function ‘TensorWrapper calcOpticalFlowSparseToDense(TensorWrapper, TensorWrapper, TensorWrapper, int, int, float, bool, float, float)’:
/tmp/luarocks_cv-scm-1-490/torch-opencv/src/optflow.cpp:54:9: error: ‘calcOpticalFlowSparseToDense’ is not a member of ‘optflow’
         optflow::calcOpticalFlowSparseToDense(
         ^
/tmp/luarocks_cv-scm-1-490/torch-opencv/src/optflow.cpp:54:9: note: suggested alternative:
/tmp/luarocks_cv-scm-1-490/torch-opencv/src/optflow.cpp:48:22: note:   ‘calcOpticalFlowSparseToDense’
 struct TensorWrapper calcOpticalFlowSparseToDense(
                      ^
/tmp/luarocks_cv-scm-1-490/torch-opencv/src/optflow.cpp:59:9: error: ‘calcOpticalFlowSparseToDense’ is not a member of ‘optflow’
         optflow::calcOpticalFlowSparseToDense(
         ^
/tmp/luarocks_cv-scm-1-490/torch-opencv/src/optflow.cpp:59:9: note: suggested alternative:
/tmp/luarocks_cv-scm-1-490/torch-opencv/src/optflow.cpp:48:22: note:   ‘calcOpticalFlowSparseToDense’
 struct TensorWrapper calcOpticalFlowSparseToDense(
                      ^
/tmp/luarocks_cv-scm-1-490/torch-opencv/src/optflow.cpp: In function ‘DenseOpticalFlowPtr createOptFlow_SparseToDense_optflow()’:
/tmp/luarocks_cv-scm-1-490/torch-opencv/src/optflow.cpp:151:32: error: ‘createOptFlow_SparseToDense’ is not a member of ‘optflow’
     return rescueObjectFromPtr(optflow::createOptFlow_SparseToDense());
                                ^
CMakeFiles/optflow.dir/build.make:54: recipe for target 'CMakeFiles/optflow.dir/src/optflow.cpp.o' failed
make[2]: *** [CMakeFiles/optflow.dir/src/optflow.cpp.o] Error 1
CMakeFiles/Makefile2:340: recipe for target 'CMakeFiles/optflow.dir/all' failed
make[1]: *** [CMakeFiles/optflow.dir/all] Error 2
Makefile:116: recipe for target 'all' failed
make: *** [all] Error 2

Error: Build error: Failed building.

BFMatcher problems

I am having problems trying to use BFMatcher. I am doing the following (similar to the flann demo)

matcher = cv.BFMatcher{cv.NORM_L2, False}
print (matcher)
local N_samples, N_features, fieldSize, N_query = 300, 2, 600, 20
local samples = torch.rand(N_samples, N_features):float() * fieldSize
local query = torch.rand(1, N_features):float() * fieldSize
local res = matcher:match{query, samples}

and I am getting

features2d/init.lua:1391: bad argument #1 to 'DescriptorMatcher_match_trainDescriptors' (cannot convert 'struct TensorWrapper' to 'struct PtrWrapper')

I readed init.lua but I don't see what I am doing wrong.

compilation on OS X

I cannot compile on Os X with brew:

$ brew install opencv
Warning: homebrew/science/opencv-2.4.12 already installed

$ make
Scanning dependencies of target Common
[ 3%] Building CXX object CMakeFiles/Common.dir/src/Common.cpp.o
In file included from /Users/eugenioculurciello/Code/Lua/torch-opencv/src/Common.cpp:1:
/Users/eugenioculurciello/Code/Lua/torch-opencv/include/Common.hpp:2:10: fatal error: 'opencv2/core.hpp' file not found

include <opencv2/core.hpp>

1 error generated.
make[2]: *** [CMakeFiles/Common.dir/src/Common.cpp.o] Error 1
make[1]: *** [CMakeFiles/Common.dir/all] Error 2
make: *** [all] Error 2

I fix that:
on os x, it is: opencv2/cora/core.hpp
then:

$ make
Scanning dependencies of target Common
[ 3%] Building CXX object CMakeFiles/Common.dir/src/Common.cpp.o
In file included from /Users/eugenioculurciello/Code/Lua/torch-opencv/src/Common.cpp:1:
/Users/eugenioculurciello/Code/Lua/torch-opencv/include/Common.hpp:146:30: error: no type named 'Moments' in namespace 'cv'
MomentsWrapper(const cv::Moments & other);
~~~~^
/Users/eugenioculurciello/Code/Lua/torch-opencv/include/Common.hpp:147:25: error: no type named 'Moments' in namespace 'cv'
inline operator cv::Moments() {
~~~~^
/Users/eugenioculurciello/Code/Lua/torch-opencv/src/Common.cpp:154:42: error: no type named 'Moments' in namespace 'cv'
MomentsWrapper::MomentsWrapper(const cv::Moments & other) {
~~~~^
3 errors generated.
make[2]: *** [CMakeFiles/Common.dir/src/Common.cpp.o] Error 1
make[1]: *** [CMakeFiles/Common.dir/all] Error 2
make: *** [all] Error 2

THis I am not sure, is this platform dependent?

imgcodecs

imread
imreadmulti
imdecode
imencode

Problems using torch-opencv

Hi.

torch-opencv looks like a good project, but I am having problems using it. I am on Ubuntu Linux 14.04 on x86_64 hardware. I have torch7 and opencv-3.0.0 installed and have been using both for some time. Up to now I have used the C++ interface to opencv, but I would like to interface to it from torch. So I downloaded torch-opencv-master.zip, and built torch-opencv with the commands:

unzip torch-opencv-master.zip
cd torch-opencv-master
mkdir build && cd build
cmake .. -DOpenCV_DIR=/usr/local/opencv-3.0.0/share/OpenCV
make

The configuration phase seemed to go ok, except for the warning about opencv_optflow:

-- The C compiler identification is GNU 4.8.4
-- The CXX compiler identification is GNU 4.8.4
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
CMake Warning at /usr/local/opencv-3.0.0/share/OpenCV/OpenCVConfig.cmake:219 (message):
opencv_optflow is required but was not found
Call Stack (most recent call first):
CMakeLists.txt:17 (FIND_PACKAGE)

-- Found Torch7 in /home/david/torch/install
-- Configuring done
-- Generating done
-- Build files have been written to: /home/david/torch/torch-opencv-master/build

The build appeared to succeed, but now I can't figure out how to tell luajit where to find torch-opencv, in other words how to get "require 'cv'" to work. I imagine that once you create a rockspec I will be able to just run "luarocks install", but meanwhile I'm not sure what symlinks I need to make and/or paths to set to connect torch's luajit to torch-opencv.

I appreciate any help you can provide.

Thanks,

-- Dave Slate

cudastereo

videoio tests

Classes:

VideoCapture
VideoWriter

cudabgsegm

ml

OMG, I shouldn't have mentioned all these functions here

ParamGrid

TrainData

StatModel

NormalBayesClassifier

ctor
predictProb

KNearest

SVM

Kernel

EM

DTrees

RTrees

Boost

ANN_MLP

LogisticRegression

optflow tests

Loading modules dynamically

(Probably) try to load modules only on demand after local cv = require 'cv' like it's done here:

https://github.com/stevedonovan/Penlight/blob/master/lua/pl/import_into.lua

Even if we decide not to use this, I still have troubles with splitting package into submodules. How to make this work without a global table?

local cv = require 'cv'
require 'cv.imgproc'

superres tests

Problem loading a few libraries; torch 7 older than 21 Oct 2015

After a successful build in Ubuntu 64 14.04.2, some libraries can be loaded e.g. highgui, but some cannot e.g. imgproc and videoio:

local cv = require 'cv'
require 'cv.imgproc' -- image processing

produces:

/home/omid/torch/install/share/lua/5.1/trepl/init.lua:363: /home/omid/torch/install/share/lua/5.1/trepl/init.lua:363: /home/omid/torch/install/share/lua/5.1/torch/init.lua:65: bad argument #2 to 'newmetatable' (parent class name or nil expected)
stack traceback:
    [C]: in function 'error'
    /home/omid/torch/install/share/lua/5.1/trepl/init.lua:363: in function 'require'
    [string "require 'cv.imgproc' -- image processing"]:1: in main chunk
    [C]: in function 'xpcall'
    /home/omid/torch/install/share/lua/5.1/trepl/init.lua:648: in function 'repl'
    ...omid/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:185: in main chunk
    [C]: at 0x00406670

Segfault: denoising demo

demo/denoising.lua

shrubb@samsung:~/Projects/opencv.torch$ th demo/denoising.lua ~/Pictures/2015-09-09/DSC_2191_o.jpg ~/Pictures/2015-09-09/DSC_2191_o.jpg 
*** Error in `/home/shrubb/Programs/Torch/install/bin/luajit': malloc(): memory corruption: 0x0000000001d1f4d0 ***
Aborted (core dumped)

photo tests

Classes:

Add LICENSE

Decide the form of creating tests

What we do need at the time being is a good code coverage by unit tests. How should we organize these tests? What will be the script to run? etc.

imgcodecs tests

imread
imreadmulti
imdecode
imencode

VideoCapture{filename}

Hi,
I've tried to run this functionality with web cam and it runs well. But I could not run it with any video file.
I tried this:

th> cap = cv.VideoCapture{filename='/home/user/MH.avi'}

th> cap:isOpened()
false

How to run torch-opencv outside of torch-opencv-master

Hi.

I have one more question related to my closed issue "Problems using torch-opencv #46". It looks like I can now run cv from the torch-opencv-master directory. However, I still have not figured out how to successfully access cv from other directories. No matter how I try to set paths, torch still has trouble finding the right ".lua" and/or ".so" files. Can you tell me exactly what I need to do so torch-opencv works when th is launched in a different directory?

Thanks.

Modify function interface style

Currently, we use this style:

function cv.GaussianBlur(t)
    local src =        assert(t.src)
    local dst =        t.dst
    local ksize =      cv.Size(assert(t.ksize))
    -- ....

You may notice that it's not very pleasant to call all the functions this way, as you'll have to either memorize all the argument names or to consult OpenCV docs right along:

cv.GaussianBlur{
    src = image, 
    dst = image, 
    ksize = {7,7}, 
    sigmaX = 1.5,
    sigmaY = 1.5
}

cv.Canny{
    image = image,
    edges = edges,
    threshold1 = 0,
    threshold2 = 30,
    apertureSize = 3
}
cv.demosaicing{_src=edges _dst=edges, code=2}
cv.imshow{winname="an image", image=edges}
cv.waitKey{delay=30}

That's why it may be a good idea to support positional arguments as well:

function cv.GaussianBlur(t)
    local src =        assert(t[1] or t.src)
    local dst =        t[2] or t.dst
    local ksize =      cv.Size(assert(t[3] or t.ksize))
    local sigmaX =     assert(t[4] or t.sigmaX)
    local sigmaY =     t[5] or t.sigmaY or 0
    local borderType = t[6] or t.borderType or cv.BORDER_DEFAULT

    -- ...
end

I suggest that we should use this style from now on. I'll write a script to modify all the previous code.

features2d

Tensor-Mat conversion

I know this great project is in its alpha/beta stage, but I decided to try it nevertheless:). Maybe I just expect too much from Tensor-Mat conversion, but it seems to me that it assumes contiguous Tensors and different dimension ordering than it is common in Torch (WHC instead of CHW).

Specifically, I want to call cv.resize(). I tried lots of variants (leading to segfaults, weird results or failed asserts in _OutputArray::create()) but this is the only thing which works for me:

local img = torch.FloatTensor(4, 128, 128)
img = cv.resize{src=img:transpose(1, 3):contiguous(), dsize={256, 256}}
img = img:transpose(1, 3):contiguous()

Can the transposition be maybe hidden inside TensorWrapper? Can't really OpenCV work with non-contiguous tensors? I'm I doing something really wrong?

visionlabs / torch-opencv Goto Github PK

torch-opencv's People

Contributors

Stargazers

Watchers

Forkers

torch-opencv's Issues

include <opencv2/bgsegm.hpp>

include <opencv2/core.hpp>

Recommend Projects

Recommend Topics

Recommend Org