Boogaloo's profile - activity

Thanks, no sign of TBB there, just the IPP, I must have seen the TBB parallel_for somewhere else... think I need a perso

I meant cv:Mat, sorry for the confusion. I remember seeing the parallel_for in use on Mat memory copy back when I used

I converted to using &refs, and all is working but only using the clone to copy. I stepped through the code, and I d

This is where I am confused because stepping through the copyTo() code it does do a memcpy to the destination texture's

Right! Sorry, that makes sense! Funny about the voodoo, when I get confused I do feel like that at times, I guess we al

I understand that in modern CPUs each core has its own PCIe channel, which is why when I tested OCV 3.1 and TBB was used

TBB not being used for CMat memory copy, just IPP I'm using 3.4.0 and enabled WITH_TBB, but stepping thro the copyTo and

CopyTo loding data I've stepped through the code, through each OpenCV 3.4.0 function, and I just can't understand why da

Has anyone come across a crash using the Intel x5-Z8300 CPU with AVX the instruction vmovq?

I've compiled OpenCV for 32-bit and it works fine on an Intel i5, but on the z8300 its crashing on this vmovq instruction.

I'm not sure the Atom CPU supports vector extensios, I´ve been looking at the Intel datasheets, but it doesn't seem clear, but it doesn't list the z8300 (Cherry Trail) as being AVX ready here: https://en.wikipedia.org/wiki/Advance...

If AVX isn't supported, should I just disable it and rebuild...?

But CMAKE should change the project settings after being run, so when I open it all in VS the settings should be correct. There is no way to do this on all the different solutions in VS, and there are scores of them so doing it manually would take a long time.

You need to sum the differences otherwise the error level will be so high you are unlikely to see a useful output. The left/right camera images will have pixels for the same objects with slightly different values due to lighting, camera noise, im perfect calibration, so the whole process is prone to error.

The first time I installed OpenCV 3.0 I did not set the check boxes for SSE2 etc., so I ran CMAKE again and checked them, after generating without errors, and SSE2 set, I loaded the project into Visual Studio Community 2015 and all of the SSE2 settings disabled.

If anyone has a solution for this, please post!

Is this function for correcting matching points on two images, or can it be used to align points on the chessboard? I can't find any examples of this function being used, and I having to wade through all the code to find out is not a good use of anyone's time.

I am looking for information on how to use correctMatches() to improve any errors in the chessboard points found during calibration.

I understand that this function determines the average epipolar line of the available points, and looks for any obvious discreancies, but I am unsure how to use the function, which requires a fundamental matrix, which I think is not calculated until after calibration.

Do I calibrate the camera using stereoCalibrate, then use the F matrix output with correctMatches, and re-enter the corrected points into stereoCalibrate again?

My advice avoid c#, its a Microsoft scam to make a commercial language. C++ is by far more advanced and more importantly, driven by community goals, and not one commercial company. If you must use c#, use C/C++ with OpenCV and interface the module to your c# code.

I´ve calibrated my cameras using the standard Calib sample, but the images are both rotated clockwise by about 12 degrees.

The cameras are quite well aligned, so I'm puzzled as to why this rotation would occur.

Is there a way to fix this?

Yes seen that, the problem is that Alalek says that compiling the wrapper is not easy. I think it has to be done with a variety of tools and is tricky. Its a pity because if conformed with the rest of OpenCV people would be able to add functionality.

Thanks! I'd like to modify and compile the functions such as icvSetCaptureProperty_FFMPEG_p etc., but there is no source code for the binary dll as far as I can see. Have now asked the author, alalek via GitHub.

In the OpenCV source ffmpeg folder there are only dll binaries, but these are for opencv versions of a wrapper.

Where can we download the source to the cv_ffmpeg, as I'd like to add functionality?

I have the same problem, in the OpenCV source ffmpeg folder only has binaries, but these are opencv versions of ffmpeg, or at least a wrapper. Where is the source to the cvffmpeg?

Isn't OpenCL version 2.0 supported by OpenCV 3.0.0?

When you create an OCL context, you can set the device type. See the create function.

Okay, so I´ve read this https://www.khronos.org/registry/cl/s... it indicated that the latency is that on first load the program has to be compiled, but that the binary can be downloaded and stored for later use, avoiding the compilation delay, which is useful, although the page does not give a link to the functions to download the compiled binary as far as I can see.

Great :| OCL has been around years and this isn't yet covered?! its sort of hard to believe. Experiments? This should just be documented, the load program function must have some details, I'm guessing that once loaded a program stays resident until deleted... is there really no specs on this?

Okay, well, would it be useful to have DirectX-DirectShow<->UMat transfers/sharing on GPU?

Given the importance of fast memory transfer to GPU from DirectShow and DirectX, it is very surprising that support for these operations, especially given the marketing of unified memory architecture, is so very poor or simply non-existent. If I had access to the damn AMD or Intel driver code I could add this in very short time, so why doesn't this get fixed?! (that's a rant! :)

I´m having similar problems using AMD GPU, support for DirectX is possibly not implemented on non-Intel devices.

Did you make any progress?

Just found this from 2013 on the AMD forum "By German on 09-Mar-2013 18:43 'Does runtime report cl_khr_dx9_media_sharing extension? You can't use an extension if runtime doesn't report it:-) Anyhow we should support DX9 interop internally, but runtime doesn't report the extension:-). There is a limitation. You have to use IDirect3DDevice9Ex device object and CL_ADAPTER_D3D9EX_KHR as adapter type. Please note, if shared_handle is NULL, then you will have double copy (acquire/release) - no direct access." - Does anyone have any more info?

Okay, cv::directx::ocl::initializeContextFromDirect3DDevice9(..) is failing to find the Khr extension for surface sharing on any installed device; I have an AMD R290 GPU. I´m wondering if AMD support this in their OpenCL driver, or if DX9 support is dropped?

Is there any updated help on this for 3.0.0? There is nothing listing on this forum, and in the web intel-only examples that are out of date, and other stuff also that seems OOD.

I've just noticed that the valid region for the rectified image is wrong, it has a vertical misalignment as you can see on the screen grab: https://www.sendspace.com/file/z79c8i

I have the stereo calibration sample working perfectly using the supplied chessboard frames, but when using my own images, that have the same VGA resolution and similar chessboard poses, the resultant rectified images are reduced to the centre area of the VGA image, i.e. the camera has zoomed out, also both are rotated 20 degrees to counterclockwise.

I suppose the R and T values from stereocalibrate are causing this, but why? My own camera-grabbed images have been made to closely match the sample images, and look almost identical.

The only obvious difference is that my chessboard is 10 x 7, and the sample's is 8 x 6 - but I did change the board parameter to 9 x 6!

I'm getting and RMS error of 0.632, Avg of 1.208

What could be causing the correctly rectified results to be scaled down, like the camera has been pulled back?

I´ve seen that there is a large latency, 0.6s, on the first call to OCL functions, and assume this is due to the function being uploaded and initialised. What is the persistency of functions once used? Can they be loaded for permanent readiness?

Are there any docs on this topic?

Did you look at the header? All three version of the struct are declared if none of the __D3D... are defined.

The structures are ONLY REMOVED when someone defines the __D3D...

I would expect to have to define what I want to use, to use it, not define what I don't want.

I do no understand you: can you explain why the controls exists in the first place, as all they do is allow selective removal of the structures, rather than the usual case of controls enabling things.

If no declarations are made, all structures are available, however, if someone declairs __d3d11_h__ them DX11 is excluded, but DX9 and 10 are available.

Sort of weird logic if you ask me!

setDevices is no longer available on 3.0, so there is no way to run-time select which GPU should be used? That would be an incredible design error given that many systems use dual or more GPUs, and sharing tasks between them is critical!

In the header there are 3 tests for __d3d11_h__ and 10 and 9 versions, but the tests only include the structures if these are NOT defined!

#if !defined(__d3d11_h__)
struct ID3D11Device;
...

Shouldn't the negate be removed, so:-

#if defined(__d3d11_h__)
struct ID3D11Device;
...

Strange, your version should be using the same compiler I imagine, so why the failure in my case? I´ll check it later and post back if I can find a solution.

With it set to V140 not v100? Did you enable SSE4 AVX POPCNT etc?

What are you compiling it with? I used service pack 1 (just released) on Visual Studio 2015 Community. The compiler still has bugs and is crashing. Sp1 fixed some crashes, but not all. I am compiling OpenCV Dev with VS 2014 without problem.

Unfortunately, even with SP1 VS 2015 compiler is still crashing compiling the Install in debug. C1001's everywhere!

They were all simple ambibuous calls, apart from a call to "round" that wasn't included; I fixed them all and so they don´t list, so the best way would be to compile the original 3.0.0-dev master with VS 2010