6 replies to this topic

[AhmedCoeia]

Hi All,

 

I'm just trying to feed a cvMat a texture that is generated by
fragment shader, there is nothing appears on the screen, I don't know
where is the problem, is this in the driver or glreadPixels.. I just
loaded a TGA Image, to a fragment shader, then textured a quad, I wanted
to feed that texture to a cvMat, so I used glReadPixesl then genereated
a new texture, and drew it on the quad, but nothing appears.

 

The setup is on OpenGLES on Exynos, Mali400 GPU
 

Kindly note that the following code is executed at each frame.

 

glEnableVertexAttribArray( userData->positionloc);
glEnableVertexAttribArray( userData->texCoordLoc);

// use Fragment Shader Texture
glActivateTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE2D, userData->baseMapTexId);


// try to store it into cvMat or GLByte
cv::Mat pixels;
glPixelStorei(GL_PACK_ALIGNMENT, (pixels.step & 3) ? 1 : 4);
glReadPixels(0, 0, 1024, 1024, GL_RGB, GL_UNSIGNED_BYTE, pixels.data);

glEnable(GL_TEXTURE_2D);
GLuint textureID;
glGenTextures(1, &textureID);
 
// Create the texture
glTexImage2D(GL_TEXTURE_2D, // Type of texture
0, // Pyramid level (for mip-mapping) - 0 is the top level
GL_RGB, // Internal colour format to convert to
1024, // Image width i.e. 640 for Kinect in standard mode
1024, // Image height i.e. 480 for Kinect in standard mode
0, // Border width in pixels (can either be 1 or 0)
GL_RGB, // Input image format (i.e. GL_RGB, GL_RGBA, GL_BGR etc.)
GL_UNSIGNED_BYTE, // Image data type
pixels.data); // The actual image data itself

// try to draw that stored texture into a quad
glActiveTexture ( textureID );
glBindTexture ( GL_TEXTURE_2D,textureID );
glDrawElements ( GL_TRIANGLES, 6, GL_UNSIGNED_SHORT, indices );
 

Edited by AhmedCoeia, 20 February 2013 - 09:27 AM.

[Brother Bob]

Don't you have to specify the size and the format of the image before passing its pointer to glReadPixels, or does the cv::Mat class default constructor allocate enough storage space to acomodate a 1024x1024 RGB image? Also, if you're creating a new texture object, then you need to set the minification filter to a non-mipmap filter or generate all mipmap levels, or you will have an incomplete texture object.

[AhmedCoeia]

Thanks for your reply.

I don't need to specify the size of the cvMat.The default constructor allocates enough space.

I will set the manification filter. Don't you see any other problem ?  I write to write glreadpixels to a file, it only writes when it is specified as GLRGBA, and doesnt work with GLRGB...

Is the shader is a problem ? Should I disable it before glreadpixesl ?

[Brother Bob]

Thanks for your reply.

I don't need to specify the size of the cvMat.The default constructor allocates enough space.

I will set the manification filter. Don't you see any other problem ?

I don't see anything obvious, no.

I write to write glreadpixels to a file, it only writes when it is specified as GLRGBA, and doesnt work with GLRGB...

Is the shader is a problem ? Should I disable it before glreadpixesl ?

The shaders do not affect glReadPixels.

[mhagain]

Check your current glReadBuffer.  Also, if you've a single-buffered context you may need a glFlush before reading (although glReadPixels implies a sync but it may be a driver bug so worth checking).

 

While we're on the subject, this is slow:

glReadPixels(0, 0, 1024, 1024, GL_RGB, GL_UNSIGNED_BYTE, pixels.data);

And this is fast:

glReadPixels(0, 0, 1024, 1024, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV, pixels.data);

The reason why is because the slow version requires a software conversion step from your native framebuffer format to the format you specify; using the same format for both avoids that conversion.

[AhmedCoeia]

Would you tell me how do I check my current glReadBuffer?  I read the docs, but didn't get it.

[mhagain]

From the link I provided:

 

Associated Gets

glGet with argument GL_READ_BUFFER

 

Follow the "glGet" link and search for GL_READ_BUFFER and we find:

 

params returns one value, a symbolic constant indicating which color buffer is selected for reading. The initial value is GL_BACK if there is a back buffer, otherwise it is GL_FRONT.

 

Because we know that the read buffer is specified by a GLenum, and because we've read the documentation we know that a GLenum is just an integer, the code for it is:

 

// ensure that our read buffer is GL_BACK
GLenum currentReadBuffer;
glGetIntegerv (GL_READ_BUFFER, &currentReadBuffer);
if (currentReadBuffer != GL_BACK) glReadBuffer (GL_BACK);

 

Incidentally, if the only reason you want to read back the data is to transfer it to a texture, consider using (preferably) an FBO or (alternatively) glCopyTexImage2D or glCopyTexSubImage2D.  Either way will put the data into a texture directly on the GPU without needing a round-trip to-and-from system memory beforehand.

 

Also, you're not calling glBindTexture before your glTexImage2D in your code.  Not to mention generating a brand-new texture object each frame.  I'd suggest after all this that you may get more benefit from jumping back to more basic tutorial work instead of trying anything too fancy right now, because your approach appears to be one of hacking-and-slashing at code rather than working to understand what's actually happening and what you actually need to do.