[❔ other question] How to convert/extract enoki.autodiff_cuda.Float32 type array from Bitmap class?

[vucenovic]

[❔ other question] How to convert/extract enoki.autodiff_cuda.Float32 type array from Bitmap class?

[merlinND]

Hello @vucenovic,

I've transferred your issue to the Mitsuba 3 repo, hopefully this is correct. If not, I strongly encourage you to upgrade, as Mitsuba 3 has many improvements and bugfixes over Mitsuba 2.

Here's a good way to do it with the Python API:

>>> b
Bitmap[
  pixel_format = rgba,
  component_format = uint8,
  size = [500, 656],
  srgb_gamma = 1,
  struct = Struct<4>[
    uint8 R; // @0, normalized, gamma
    uint8 G; // @1, normalized, gamma
    uint8 B; // @2, normalized, gamma
    uint8 A; // @3, normalized, alpha
  ],
  data = [ 1.25 MiB of image data ]
]
>>> mi.TensorXf(b)
TensorXf(shape=(656, 500, 4))
>>> mi.TensorXf(b).array
[0.0, 0.0, 0.0, 0.0, 0.0, .. 1311990 skipped .., 0.0, 0.0, 0.0, 0.0, 0.0]
>>>

On the last line, you can see that the tensor is just a wrapper around a Float object. We recommend keeping the TensorXf though, since it maintains the shape of the data as well.

[vucenovic]

Thanks for the quick response! Sadly I am not able to upgrade for the time being, would this also work on Mitsuba2? And could you maybe add the required imports to the solution (what is 'mi')?

[merlinND]

Here would be an option for Mitsuba 2:

>>> import numpy as np
>>> import mitsuba
>>> mitsuba.set_variant('cuda_ad_rgb')
>>> b = mitsuba.core.Bitmap('path/to/image.exr')
>>> mitsuba.core.Float(np.array(b).ravel())
KeyboardInterrupt
>>> mitsuba.core.Float(np.array(b).ravel())
[0.0, 0.0, 0.0, 0.0, 0.0, .. 1311990 skipped .., 0.0, 0.0, 0.0, 0.0, 0.0]

mi is just the recommended alias we use in Mitsuba 3.
Unfortunately, we most likely won't have the bandwidth to provide support for Mitsuba 2. The migration shouldn't be too hard (majority of changes are components having moved around and some renames), and would be well worth it.

[vucenovic]

Thank you! I will look into Mitsuba3 and the migration process soon :)

Kind regards