Directional lights are 4800x dimmer than point lights and spot lights

[fintelia]

Bevy version

2ec38d1 (but also happens on 0.10.x and probably others).

What you did

Delete the environment map lighting from the PBR example, then compare point light and directional light strengths.

The point light is sqrt(50^2 + 50^2 + 50^2) ≈ 86 meters from the origin and has a strength of 600K lumens. Which means 600,000 / (4 * PI * 86^2) ≈ 6.5 lux reaching the origin.

Point light

Same strength directional light

commands.spawn(DirectionalLightBundle {
    transform: Transform::from_xyz(50.0, 50.0, 50.0).looking_at(Vec3::ZERO, Vec3::Y),
    directional_light: DirectionalLight {
        illuminance: 6.5,
        ..default()
    },
    ..default()
});

4800x scaled directional light

commands.spawn(DirectionalLightBundle {
    transform: Transform::from_xyz(50.0, 50.0, 50.0).looking_at(Vec3::ZERO, Vec3::Y),
    directional_light: DirectionalLight {
        illuminance: 6.5 * 4800.0,
        ..default()
    },
    ..default()
});

Additional information

The bug is caused by this code which dims the strengths of directional lights by 4800x, but leaves point lights and spotlights unchanged:

bevy/crates/bevy_pbr/src/render/light.rs

Lines 857 to 867 in 2ec38d1

	// convert from illuminance (lux) to candelas
	//
	// exposure is hard coded at the moment but should be replaced
	// by values coming from the camera
	// see: https://google.github.io/filament/Filament.html#imagingpipeline/physicallybasedcamera/exposuresettings
	const APERTURE: f32 = 4.0;
	const SHUTTER_SPEED: f32 = 1.0 / 250.0;
	const SENSITIVITY: f32 = 100.0;
	let ev100 = f32::log2(APERTURE * APERTURE / SHUTTER_SPEED) - f32::log2(SENSITIVITY / 100.0);
	let exposure = 1.0 / (f32::powf(2.0, ev100) * 1.2);
	let intensity = light.illuminance * exposure;

[alice-i-cecile]

This will be a silly migration. Better to fix it up sooner than later.

[benny-n]

Hi, I'd like to take this one, if possible. I think I have a rough idea of what to change to solve it.
Let me know if there's anything else I should know before getting started.

[CptPotato]

I think this is because of a fundamental difference between point/spot lights (which have a spatial position and a 1/d² falloff) and directional lights that are infinitely far away and have no distance attenuation.

With spot and point lights you can assign them a value like 1500 lumens that corresponds to the amount of light emitted - which does not work for directional lights. They're using illuminance (lux), which refers to the amount of light received instead. So you're essentially using completely different units for these types of lights.

If I remember correctly the PBR implementations somewhat follows the Filament docs. There is some more information on light units there.

That being said, maybe one of the people who worked on the PBR implementation can chime in and confirm that the current behavior is correct.

[fintelia]

This is not caused by the inverse square falloff. In my post, I use the amount of light emitted by the point light together with the distance to calculate the amount of light received at the origin. Then I construct a directional light that produces the same illuminance.

Rather, this is caused by not following the Filament docs. Specifically, section 5.2.6 says that you have to pre-expose both directional and point/spot lights but Bevy is only pre-exposing directional lights while forgetting to do the same for point/spot lights.

[superdump]

It seems like we could implement a quick fix that uses hard-coded exposure settings, but also that using per-camera defined exposure settings should be pretty straightforward to implement anyway?

[fintelia]

There's currently Camera3dBundle::color_grading::exposure, but that's an exposure value offset which means that it is intended to be subtracted from the calculated automatic exposure. The main problem with using it as a fixed exposure setting directly is that it is negated (if you want a scene ev=15, then you'd set exposure=-15)