What is the best way to solve this?
glPixelZoom() does not work because I am not using glReadPixels() and friends (using FBO’s with glCopyTexSubImage2D() as a fallback).
I don’t want to mess with the projection matrix unless that is the only solution. Such a system means setting a flag to generically Y-invert the projection matrix on render, reversing the culling, etc.
A system prone to error, I think.
I am looking for a way in OpenGL to either flip my image on render or to quickly Y-invert the FBO after render.
Regards,
Yogurt Emperor
Rendering to Textures is Up-Side Down
"A system prone to error, I think."
Why?
You're only going to have two routines; one which sets up FBO rendering and one which sets up main display rendering and you call the appropriate one when needed. They aren't going to be very long; there shouldn't really be a lot of room for error.
Why?
You're only going to have two routines; one which sets up FBO rendering and one which sets up main display rendering and you call the appropriate one when needed. They aren't going to be very long; there shouldn't really be a lot of room for error.
In my opinion, the correct solution is, as obvious as it may seem, to do everything correct from the beginning and play by OpenGL's natural rules and conventions.
The typical problem is that you use a texture coordinate space with origin in the upper left corner, while OpenGL typically favors the lower left origin convention. The result of mixing these up is a vertically flipped texture. There are many ways to handle it, but play with OpenGL's convention, instead of prying it into your convention, and everything will come out not only correct, but also naturally and with little problems.
The actual problem is likely that you don't really know and have control of where the origin of your images are. Since you have a 50% chance to get it right the first time (vertical mirroring is typically the only issue you encounter) and the solution when its wrong is simple, you just go ahead and don't think more about it anymore once it's "working".
There are many pieces that works together to orient the image properly, like the order of the source data, the direction of the texture coordinate axes relative the object, the direction of the vertex coordinate axes relative the screen, and if you don't have full control and knowledge of each and every one of them, you will end up with these kinds of problems sooner or later. The solution may be in any of those pieces that orients an image, so your best option is to learn and do it right all the way.
The typical problem is that you use a texture coordinate space with origin in the upper left corner, while OpenGL typically favors the lower left origin convention. The result of mixing these up is a vertically flipped texture. There are many ways to handle it, but play with OpenGL's convention, instead of prying it into your convention, and everything will come out not only correct, but also naturally and with little problems.
The actual problem is likely that you don't really know and have control of where the origin of your images are. Since you have a 50% chance to get it right the first time (vertical mirroring is typically the only issue you encounter) and the solution when its wrong is simple, you just go ahead and don't think more about it anymore once it's "working".
There are many pieces that works together to orient the image properly, like the order of the source data, the direction of the texture coordinate axes relative the object, the direction of the vertex coordinate axes relative the screen, and if you don't have full control and knowledge of each and every one of them, you will end up with these kinds of problems sooner or later. The solution may be in any of those pieces that orients an image, so your best option is to learn and do it right all the way.
Author
Katie, I have enough control over my system that I could indeed implement it in a safe and organized way. I guess I should have said it would feel hacky to do it that way, but I am prepared to do it if necessary.
Brother Bob, I guess I should have mentioned my reason for wanting to do this, because I guess I did come off as sounding like a newbie who encounters one thing acting strangely and then just hacking it to get it to work the way he or she wants.
I do know why it is that way and I have a solid reason for wanting to change it in this particular case.
My engine supports DirectX (9 and above) and they render-to-texture without Y-inverting the image (because they assume upper-left origin of the frame buffer).
I know why they are doing what they are doing, but I do need the same series of function calls to yield the same result. That is imperative.
Now, I can choose to go either way, but the DirectX way is more intuitive for the (somewhat specialized) feature I am trying to implement.
This type of render-to-texture is meant for visual use, not internal workings such as generating shadow maps or something else that is a small part of a larger operation. Typical use would be to take a screenshot, apply the image to a bunch of triangles that initially cover the whole screen, and have the triangles fall off like broken glass for a nifty-puff menu effect.
The effect should work just as nicely if the image comes from a file instead of a screenshot, so you can understand my desire to go with the DirectX way here.
Also, flipped textures do have a use, especially for mapping onto models. I will have a flag set to optionally not flip the texture in OpenGL, and to flip it in DirectX. But either way, I need a way to flip it in both OpenGL and DirectX.
I am guessing that this will be through the projection matrix. Drat it.
Thank you,
Yogurt Emperor
Brother Bob, I guess I should have mentioned my reason for wanting to do this, because I guess I did come off as sounding like a newbie who encounters one thing acting strangely and then just hacking it to get it to work the way he or she wants.
I do know why it is that way and I have a solid reason for wanting to change it in this particular case.
My engine supports DirectX (9 and above) and they render-to-texture without Y-inverting the image (because they assume upper-left origin of the frame buffer).
I know why they are doing what they are doing, but I do need the same series of function calls to yield the same result. That is imperative.
Now, I can choose to go either way, but the DirectX way is more intuitive for the (somewhat specialized) feature I am trying to implement.
This type of render-to-texture is meant for visual use, not internal workings such as generating shadow maps or something else that is a small part of a larger operation. Typical use would be to take a screenshot, apply the image to a bunch of triangles that initially cover the whole screen, and have the triangles fall off like broken glass for a nifty-puff menu effect.
The effect should work just as nicely if the image comes from a file instead of a screenshot, so you can understand my desire to go with the DirectX way here.
Also, flipped textures do have a use, especially for mapping onto models. I will have a flag set to optionally not flip the texture in OpenGL, and to flip it in DirectX. But either way, I need a way to flip it in both OpenGL and DirectX.
I am guessing that this will be through the projection matrix. Drat it.
Thank you,
Yogurt Emperor
Do this before drawing (pick the appropriate call)::
glOrtho (x, width, height, y, zn, zf); // top-left is the origin
glOrtho (x, width, y, height, zn, zf); // bottom-left is the origin
Author
Thank you for the suggestions but I need this to be transparent to the user, work under any projection (orthogonal and perspective), and require no change in the user’s code to get the same result in both Direct3D and OpenGL.
I have organized my renderer so that one function is used to make the final render.
In this function I combine matrices to create the model-view, model-view-projection, etc.
I have added a flag the user can change to Y-invert the render(s). If this flag is set, I, for the duration of that call to render, Y-invert the projection matrix and invert the cull mode, then switch them back when the render function exits.
At first I thought this would be hacky but it works out as an engine feature and gives me flexibility since it also works for Direct3D.
I added a flag to the render-target textures so the user can decide if the resulting texture should be right-side up or up-side down, and the resulting textures are consistent between API’s.
Thank you for all the input.
Yogurt Emperor
I have organized my renderer so that one function is used to make the final render.
In this function I combine matrices to create the model-view, model-view-projection, etc.
I have added a flag the user can change to Y-invert the render(s). If this flag is set, I, for the duration of that call to render, Y-invert the projection matrix and invert the cull mode, then switch them back when the render function exits.
At first I thought this would be hacky but it works out as an engine feature and gives me flexibility since it also works for Direct3D.
I added a flag to the render-target textures so the user can decide if the resulting texture should be right-side up or up-side down, and the resulting textures are consistent between API’s.
Thank you for all the input.
Yogurt Emperor
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
