Batch Sensor working with PTracer

[Ghiraa]

Is there a method to utilize a batch sensor with the "ptracer" integrator in Mitsuba without creating a custom sensor? If this isn't possible, could you recommend any examples of custom sensors for guidance? Similarly, I'm interested in developing a custom integrator similar to "ptracer" that also considers participating media, akin to "volpath". As I'm new to Mitsuba, any advice or direction on where to find relevant resources or examples would be greatly appreciated. Thanks!

[njroussel]

Hi @Ghiraa

I believe ptracer should work with batch sensor. Are you getting an error message?

As for a custom integrator, in addition to the ptracer I would recommend having a look at prbvolpath.sample() and prb_projective.sample_importance(). The former should help you with the Python interface of the medium sampling and the latter with the light tracing style of rendering to connect your path to a sensor.
Other than that, PBRT should guide you through the actual volume rendering and light tracing theory if you're unfamiliar with it.

[Ghiraa]

I can't run the code above, my kernel keeps dying for some reason.

[merlinND]

@Ghiraa If you are encountering a crash, please run the code as a Python script in the terminal and report which error is printed before the crash.

[Ghiraa]

Hi, sorry for coming back so late to this post. I ran it as a python script and the issue is that I could not use the "'cuda_ad_rgb'", but after changing it it worked perfectly fine. I still wonder what's issue with my code though, here is how I create them:

def load_sensor(r, phi, theta):
    # Apply two rotations to convert from spherical coordinates to world 3D coordinates.
    origin = T.rotate([0, 0, 1], phi).rotate([0, 1, 0], theta) @ mi.ScalarPoint3f([0, 0, r])

    return {
        'type': 'perspective',
        'fov': 39.3077,
        'to_world': T.look_at(
            origin=origin,
            target=[0, 0, 0],
            up=[0, 0, 1]
        ),
        'sampler': {
            'type': 'independent',
            'sample_count': 16
        },
        'film': {
            'type': 'hdrfilm',
            'width': 256,
            'height': 256,
            'rfilter': {
                'type': 'tent',
            },
            'pixel_format': 'rgb',
        },
    }

sensor_description = {
    'type': 'batch',
    'sampler': {
            'type': 'independent',
            'sample_count': 16
    },
    'film': {
        'type': 'hdrfilm',
        'width': 6 * 256,
        'height': 254,
        'rfilter': {
            'type': 'tent',
        },
        'pixel_format': 'rgb',
    },
}

sensor_count = 6

radius = 6
phis = [60.0 * i for i in range(sensor_count)]
theta = 60.0
for i, phi in enumerate(phis):
    sensor_description[f"persp_{i}"] = load_sensor(radius, phi, theta)

sensorr = mi.load_dict(sensor_description)
image = mi.render(scene, spp=16, sensor=sensorr)

sensors = [mi.load_dict(load_sensor(radius, phi, theta)) for phi in phis]

[merlinND]

Glad to hear that you got it working. If it works, then what do you mean by "I still wonder what's issue with my code though"?
What is not going as expected?

[njroussel]

@Ghiraa There's a typo in your code:

'film': {
        'type': 'hdrfilm',
        'width': 6 * 256,
        'height': 254, # <--- HERE
        'rfilter': {
            'type': 'tent',
        },
        'pixel_format': 'rgb',
    },

The height should be 256. I checked the source code, we never actually check that the film's height is consistent with the "sub"-sensors'.