FBO for drawing / modify an image

OpenGL Development
#1

Hello all,

For simplify, i have an Image with

  • A java ByteBuffer containing pixels
  • A render context giving an OpenGL Texture ID (if this image was already used by an OpenGL process)

So, now i want to be able to render on it and get back pixels buffer.

I would do a things like it (on general idea)

  • Bind a texture from this image if it’s not already done
  • Generate an FBO
  • Bind the FBO to the texture (or draw the texture on the FBO)
  • Bind the FBO
  • *** draw ***
  • delete the FBO (i assume that the result is directly on the texture ???)

So after all, i want to mark the image as “Dirty” that mean the ByteBuffer is not up to date.
If in java i try a getBuffer(), i want to be able:

  • Get Back pixels buffer from the texture

So, i’m newbie in OpenGL, and i’m not sure that this procedure can be done as is.
Is a good way ? i don’t sure that FBO can be used like it ?
So if you can advise me on how to achieve this, it will be really usefull.

For more explanation, I would be able to do an API for drawing on an image in the same way if i draw on the screen.

Thanks in advance.
Sébastien.

#2

Yes, it´s possible.
Something like this (untested code):


int fbo = glGenFramebuffers();
glBindFramebuffer(GL_FRAMEBUFFER, fbo);
glFramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, textureID, 0);

glViewport(0, 0, texWidth, texHeight);

//Draw to texture

glBindFramebuffer(GL_FRAMEBUFFER, 0); //Unbind FBO
glDeleteFramebuffers(fbo);

glGetTexImage(GL_TEXTURE_2D, 0, GL_RGB, GL_UNSIGNED_BYTES, byteBuffer);

//byteBuffer now contains the image data of the texture

This is slow as hell though. You could use it to generate static images at load time, but doing this every frame is a very bad idea since glGetTexImage() forces the GPU to finish rendering to the texture before it returns which stalls the CPU while waiting for it and then the GPU after it finishes since it has no more work.

#3

Thanks a lot, i will test if it’s work.
But yes, it’s intended to create image and reuse it as is but not intended to modify image each frame rendering.

#4

Based on you example, i try this:

Initialisation of the FBO:


   frameBufferId = GL30.glGenFramebuffers();
		
    GL30.glBindFramebuffer( GL30.GL_FRAMEBUFFER, frameBufferId);
    GL30.glFramebufferTexture2D(GL30.GL_FRAMEBUFFER, GL30.GL_COLOR_ATTACHMENT0, GL.GL_TEXTURE_2D, imageContext.getTextureId() , 0);
    GL11.glViewport(0 , 0 , image.getWidth(), image.getHeight() );

When a drawing operation occurs on my image context, if it’s underlying FBO is not the current frame buffer, i do this:


    GL30.glBindFramebuffer(GL30.GL_FRAMEBUFFER, frameBufferId);
    GL11.glViewport(0 , 0 , image.getWidth(), image.getHeight() );

When drawaing operations are done on the image context, i close the underlyng FBO


        // returns on windows-system FBO
   	GL30.glBindFramebuffer(GL30.GL_FRAMEBUFFER, 0);
		
	// delete the FBO
	GL30.glDeleteFramebuffers(this.frameBufferId);
	this.frameBufferId = -1;

So after this sequence, i draw the texture on a 2D QUAD (i use an ortho projection since i work only in 2D)
but the texture remains unmodified… i don’t find what is wrong…

#5

it need a call of GL11().glBindTexture(GL.GL_TEXTURE_2D, 0) before to bind the texture to the FBO

#6

Ideally you should just initialize a single FBO that covers all sizes and reuse it as needed, rather than creating/deleting many FBOs throughout your game.

You should also do a glCheckFramebufferStatus to be safe, and probably unbind the frame buffer after creating it (unless you plan on using it immediately after creation). Don’t forget to reset your viewport to your game’s size after unbinding the FBO.

Rendering to a texture looks like this (pseduo-code):

fbo.bind(); 
//now we are rendering to the FBO's color texture
//draw our stuff here...
fbo.unbind();

//now we are rendering to the screen... so we need to draw the FBO's color texture
drawTex( fbo.getTexture() );

Another thing… You might need to reset your glOrtho/modelview/etc. when entering the FBO, i.e. via a matrix push/pop.

#7

Thanks, i try to go in this way.

when i init my app, i do this

  
   GL11.glMatrixMode(GL.GL_PROJECTION);
   GL11.glLoadIdentity();
   GL11.glOrtho(0, appWidth ,  appHeight , 0, 1, -1);
   GL11.glMatrixMode(GL.GL_MODELVIEW);
   GL11.glEnable( GL.GL_TEXTURE_2D );
   GL11.glViewport(0 , 0 , getWidth() , getHeight() );

when i bind the fbo, i can do this:

  
   GL11.glPushMatrix();
   GL30.glBindFramebuffer(GL30.GL_FRAMEBUFFER, frameBufferId);
   GL11.glViewport(0 , 0 , image.getWidth() , image.getHeight() );

When i unbind the fbo, i do this

   
   GL30.glBindFramebuffer(GL30.GL_FRAMEBUFFER, 0);
   GL11.glPopMatrix();
   GL11.glViewport(0 , 0 , appWidth,  appHeight);

I have with this strange result (coordinate and scale issue that let means that i have problems with view context)
What can i miss ?

#8

Try re-entering ortho mode. Also push some attributes like GL_VIEWPORT_BIT and GL_TRANSFORM_BIT via glPushAttrib.

Here’s my little FBO wrapper if you’re interested. It uses GL_EXT_framebuffer_object since OpenGL 3.0 is still a bit new (and my Mac only supports GL 2.1).

http://www.java-gaming.org/?action=pastebin&id=79

#9

Finally this works for me :

FBO init:

   
   // we muse ensure that before that, the textureId is not bind
   frameBufferId = GL30.glGenFramebuffers();
   GL30.glBindFramebuffer(GL30.GL_FRAMEBUFFER, frameBufferId);
   GL30.glFramebufferTexture2D(GL30.GL_FRAMEBUFFER, GL30.GL_COLOR_ATTACHMENT0, GL.GL_TEXTURE_2D, textureId , 0);
   GL30.glBindFramebuffer(GL30.GL_FRAMEBUFFER, 0);

Graphics Context bind (for drawing on the image):

  
  originalViewport.clear();                                             // direct IntBuffer in native bytes order
  GL11.glGetInteger( GL.GL_VIEWPORT , originalViewport ); // backup the current viewport
    	
  GL11.glMatrixMode( GL.GL_PROJECTION );
  GL11.glPushMatrix();
    	
  GL11.glLoadIdentity();
  GL11.glOrtho(0, image.getWidth() , 0, image.getHeight() , 1, -1);
  GL11.glMatrixMode( GL.GL_MODELVIEW );
  GL11.glPushMatrix();
    	
  GL30.glBindFramebuffer(GL30.GL_FRAMEBUFFER, frameBufferId);
  GL11.glViewport(0, 0, imageWidth , imageHeight );

And when i need to push back into the window context :

   
   GL11().glMatrixMode( GL.GL_MODELVIEW );
   GL11().glPopMatrix();
   GL11().glMatrixMode( GL.GL_PROJECTION );
   GL11().glPopMatrix();
        
   GL30.glBindFramebuffer( GL30.GL_FRAMEBUFFER, 0);
   GL11.glViewport( originalViewport.get(0) , originalViewport.get(1) , originalViewport.get(2) , originalViewport.get(3) );

And the destroy part:

   
   GL30.glDeleteFramebuffers(frameBufferId);
   frameBufferId = -1;

I’m not really happy with this current design, in my opinion, it’s to my “window” class to restore it’s original viewport, but i will see after how i can enhance this.
Thank’s for all, your wrapper helps me to write mine.

#10

You can push the viewport attribute (glPushAttrib(GL_VIEWsomething) and then glPopAttrib(GL_VIEWsomething)) to save the current viewport.

#11

Thanks, i don’t know this, i will try this evening

#12 

glPushAttrib(GL_VIEWPORT_BIT);

glPopAttrib(GL_VIEWPORT_BIT);
#13

It works fine for now.

FOr add deeper in my tools, in can draw on a specific region on a texture.
it becomes when we create a “subimage” from another one at the API level.
i must in this case let it drawing only in the texture that corresponds to it’s subimage.
For now, i used a translatef + scissor like it:

    
     public void bind(GLContext ctx) {
		// set rendering destination to FBO
    	ImageContext imageContext = (ImageContext)image.getContext();
		
    	ctx.getGL11().glMatrixMode( GL.GL_PROJECTION );
    	ctx.getGL11().glPushMatrix();
    	
    	ctx.getGL11().glLoadIdentity();
    	ctx.getGL11().glOrtho( 0 , imageContext.getTextureWidth() , 0 , imageContext.getTextureHeight() , 1, -1);
    	ctx.getGL11().glMatrixMode( GL.GL_MODELVIEW );
    	ctx.getGL11().glPushMatrix();
    	
    	GL30.glBindFramebuffer(GL30.GL_FRAMEBUFFER, frameBufferId);
    	super.bind(ctx); // push all attrib-bits + all client-attrib-bits
    	ctx.getGL11().glViewport( 0 , 0 , imageContext.getTextureWidth() , imageContext.getTextureHeight() );
    	
    	ctx.getGL11().glTranslatef( imageContext.getX() , imageContext.getY() , 0f );
    	ctx.getGL11().glEnable( GL.GL_SCISSOR_TEST );
    	ctx.getGL11().glScissor( imageContext.getX() , imageContext.getY() , imageContext.getWidth() , imageContext.getHeight() );
    }

I would know if the viewport reset before translate call ? because here i explain a translate from my viewport coordonate reardless previous translatef that may be done before.
It seems to work well, bu i want to be sure.
Thanks in advance.