No longer able to retrieve depth texture from buffer about panda3d HOT 6 CLOSED

Thanks, this is fixed now, I believe. The format-choosing code had been changed to set the format to F_depth_component24, but the copy-to-RAM code in the GL code didn't understand what component type to use for F_depth_component24.

from panda3d.

Thank you!
The code snippet works again now, but I have to set the number of depth bits to 16; when it is set to 24, the depth values are all zero (when checking the pixel color of the PNMImage I store the depth texture into).

Consider the following code:

from panda3d.core import *
from direct.showbase.ShowBase import ShowBase

loadPrcFileData("", "depth-bits 24")


class MyApp(ShowBase):

    def __init__(self):

        ShowBase.__init__(self)

        depth_tex = Texture("depth_texture")
        depth_tex.set_format(Texture.F_depth_component)
        fbp = FrameBufferProperties()
        fbp.set_depth_bits(24)
        buffer = self.win.make_texture_buffer("buffer", 1, 1, depth_tex, to_ram=True, fbp=fbp)
        self.graphicsEngine.render_frame()
        img = PNMImage(1, 1)
        depth_tex.store(img)
        pixel = img.get_xel(0, 0)
        print("Pixel:", pixel)


app = MyApp()
app.run()

The output is:

Pixel: LVecBase3f(0, 0, 0)

but when I do fbp.set_depth_bits(16) I get the expected result:

Pixel: LVecBase3f(0, 0, 1)

from panda3d.

That's because PNMImage only supports up to 16 bits per component, and the current code downloads 24-bit-depth buffers into the 32-bit T_unsigned_int component type.

There's no good reason why Texture.store doesn't work for T_unsigned_int textures. I will check in a fix for that shortly. However, do you remember what component_type the data was for your code in 1.9?

from panda3d.

There's no good reason why Texture.store doesn't work for T_unsigned_int textures. I will check in a fix for that shortly.

Much appreciated!

However, do you remember what component_type the data was for your code in 1.9?

Yes, that was T_float.

On a related note, it seems that the TexturePeeker class isn't compatible with depth textures; would it be worth it to fix this (if that's even possible), in case it gave a better performance when continually checking pixel values, compared to storing the texture in a PNMImage every frame?
If you find it worthwile, I would be willing to file a new issue report for this.

P.S. (Off-topic) In case you missed it, I commented on issue #189 some time ago.

from panda3d.

OK, now 24-bit depth is converted to T_float again (e4e24ee). Also, TexturePeeker now supports both unsigned integer and floating-point textures (348b1c3). And store() is now implemented for T_unsigned_int (9dcfcbf).

Indeed, TexturePeeker is a faster way to access texture data. In fact, if you try to store a floating-point texture into a PNMImage, it will first store it into a PfmFile (the floating-point equivalent for PNMImage) and convert that into a PNMImage, so it is extra redundant. TexturePeeker is even faster because it does no extra copy or conversion at all.

I did not forget about #189, I will look into that when I find some time.

from panda3d.

Thank you very much for all those fixes and improvements! Just tested my code snippet again and everything works fine now.

Good to know about the efficiency of TexturePeeker; thanks again for extending its usefulness!

from panda3d.