Bitmap conversion changes rendered values beyond clipping. Why?

[AZaitzeff]

import drjit as dr
import mitsuba as mi
import matplotlib.pyplot as plt
import numpy as np
mi.set_variant("scalar_rgb")
scene = mi.load_file('../scenes/cbox.xml')
image_ref = mi.render(scene, spp=1024)
plt.imshow(image_ref)

bitmap=mi.util.convert_to_bitmap(image_ref)
bitmap

bit_values=mi.TensorXf(bitmap).numpy().flatten()
render_values=image_ref.numpy().flatten()
plt.scatter(render_values,bit_values)
plt.xlim(0,1)
plt.ylim(0,255)

plt.imshow(np.sqrt(np.clip(image_ref,0,1)))

So it seems that in addition to clipping, convert_to_bitmap takes the square root (or something close to it) of the RGB values. Why does convert to bitmap change the RGB values in this way and is this documented somewhere?

[njroussel]

Hi @AZaitzeff

Here's the source code of convert_to_bitmap: source. The documentation is also in our API reference.

I believe what you are referencing here is the sRGB gamma ramp/correction. We don't have a page about this topic specifically, but we do have pointers on how to enable/disable it in Mitsuba this guide.
In short, we need a linear color space to during rendering but most likely you'll want a sRGB color values when displaying the rendered image.

Also convert_to_bitmap is a convenience function. As you can see from its source code, it's only a couple of lines. You can use the full BItmap constructor and its methods if you wish to produce a different result.