Python binbings crash while printing differatiable parameters

[mitsuba-renderer]

Python binbings crash while printing differatiable parameters

[Speierers]

First of all, you are using mitsuba.set_variant('gpu_autodiff_spectral'), meaning that red.reflectance.values is not a Color3f here but a set of values (see regular.cpp).

Could you try again with gpu_autodiff_rgb (and use ``red.reflectance.value` then)?

[Speierers]

I should work with spectral rendering. You will need to provide spectral values instead of RGB here tho.

[Speierers]

Thanks for reporting this!
It should now be fixed (d066893).

[alexus37]

Wow, thanks for the quick fix! Love the spirit!

[garethwalkom]

I'm getting a similar error when using the example at https://mitsuba2.readthedocs.io/en/latest/src/inverse_rendering/pytorch.html:

>>> import enoki as ek
>>> import mitsuba
>>> mitsuba.set_variant('gpu_autodiff_spectral')
>>>
>>> from mitsuba.core import Thread, Vector3f
>>> from mitsuba.core.xml import load_file
>>> from mitsuba.python.util import traverse
>>> from mitsuba.python.autodiff import render_torch, write_bitmap
>>> import torch
>>> import time
>>>
>>> Thread.thread().file_resolver().append('cbox')
>>> scene = load_file('../resources/data/scenes/cbox/cbox.xml')
2020-04-02 00:34:07 INFO  main  [xml.cpp:1129] Loading XML file "..\resources\data\scenes\cbox\cbox.xml" ..
2020-04-02 00:34:07 INFO  main  [xml.cpp:1130] Using variant "gpu_autodiff_spectral"
2020-04-02 00:34:07 INFO  main  [PluginManager] Loading plugin "plugins\regular.dll" ..
2020-04-02 00:34:07 INFO  main  [PluginManager] Loading plugin "plugins\path.dll" ..
2020-04-02 00:34:07 INFO  main  [PluginManager] Loading plugin "plugins\independent.dll" ..
2020-04-02 00:34:07 INFO  main  [PluginManager] Loading plugin "plugins\gaussian.dll" ..
2020-04-02 00:34:07 INFO  main  [PluginManager] Loading plugin "plugins\hdrfilm.dll" ..
2020-04-02 00:34:07 INFO  main  [PluginManager] Loading plugin "plugins\perspective.dll" ..
2020-04-02 00:34:07 INFO  main  [PluginManager] Loading plugin "plugins\diffuse.dll" ..
2020-04-02 00:34:07 INFO  main  [PluginManager] Loading plugin "plugins\area.dll" ..
2020-04-02 00:34:07 INFO  main  [PluginManager] Loading plugin "plugins\d65.dll" ..
2020-04-02 00:34:07 INFO  main  [PluginManager] Loading plugin "plugins\obj.dll" ..
2020-04-02 00:34:07 INFO  main  [Scene] Validating and building scene in OptiX.
>>> params = traverse(scene)
>>> params.keep(['red.reflectance.value'])
>>> param_ref = params['red.reflectance.value'].torch()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\u0120819\mitsuba2\dist\python\mitsuba\python\util.py", line 41, in __getitem__
    return self.get_property(*(self.properties[key]))
KeyError: 'red.reflectance.value'
>>> print(params)
ParameterMap[
]

I also haven't changed anything from the example scene cbox.xml.

I'm unsure how to fix this, but I have pulled and built the gpu_autodiff_spectral variant with the latest mitsuba2 release.

[Speierers]

This should be fixed now: bffd685
Thanks!

Also is should be

params.keep(['red.reflectance.values'])

as you are in spectral mode.

[garethwalkom]

@Speierers thanks! That did the job.

Also is should be

params.keep(['red.reflectance.values'])
as you are in spectral mode.

Is the above included in the documentation? I couldn't see it when following https://mitsuba2.readthedocs.io/en/latest/src/inverse_rendering/pytorch.html

A few lines down, I now have the following error:

>>> param_ref = params['red.reflectance.values'].torch()
>>> print(param_ref)
tensor([0.0400, 0.0460, 0.0480, 0.0530, 0.0490, 0.0500, 0.0530, 0.0550, 0.0570,
        0.0560, 0.0590, 0.0570, 0.0610, 0.0610, 0.0600, 0.0620, 0.0620, 0.0620,
        0.0610, 0.0620, 0.0600, 0.0590, 0.0570, 0.0580, 0.0580, 0.0580, 0.0560,
        0.0550, 0.0560, 0.0590, 0.0570, 0.0550, 0.0590, 0.0590, 0.0580, 0.0590,
        0.0610, 0.0610, 0.0630, 0.0630, 0.0670, 0.0680, 0.0720, 0.0800, 0.0900,
        0.0990, 0.1240, 0.1540, 0.1920, 0.2550, 0.2870, 0.3490, 0.4020, 0.4430,
        0.4870, 0.5130, 0.5580, 0.5840, 0.6200, 0.6060, 0.6090, 0.6510, 0.6120,
        0.6100, 0.6500, 0.6380, 0.6270, 0.6200, 0.6300, 0.6280, 0.6420, 0.6390,
        0.6570, 0.6390, 0.6350, 0.6420], device='cuda:0')
>>> image_ref = render_torch(scene, spp=8)
>>> crop_size = scene.sensors()[0].film().crop_size()
>>> write_bitmap('out.png', image_ref, crop_size)
>>> params['red.reflectance.values'] = [.9, .9, .9]
>>> params.update()
>>> params_torch = params.torch()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "C:\Users\u0120819\mitsuba2\dist\python\mitsuba\python\util.py", line 95, in torch
    return {k: v.torch().requires_grad_() for k, v in self.items()}
  File "C:\Users\u0120819\mitsuba2\dist\python\mitsuba\python\util.py", line 95, in <dictcomp>
    return {k: v.torch().requires_grad_() for k, v in self.items()}
AttributeError: 'float' object has no attribute 'torch'

Thanks.

[merlinND]

Also I think that on this line, you might need to use an Enoki type rather than a raw list?

params['red.reflectance.values'] = [.9, .9, .9]

[Speierers]

Also I think that on this line, you might need to use an Enoki type rather than a raw list?

params['red.reflectance.values'] = [.9, .9, .9]

Not sure about this. This is how it is done in the tuto. I think the implicit cast can figure it out.

[garethwalkom]

Could you try again with gpu_autodiff_rgb (and use ``red.reflectance.value` then)?

This worked, thank you! Pretty cool. Does it not work with spectral rendering?