AI Glossary / Computer vision and multimodal AI / Intermediate

Light transport

Definition

Light transport describes how light moves through a scene and interacts with surfaces and media before reaching a sensor or observer. It includes emission, reflection, transmission, absorption, and scattering. Computer graphics models these processes to render images, while computer vision may reason in the opposite direction to estimate scene properties from captured appearance.

What it means

A surface's appearance depends on more than its own color. It also depends on illumination, geometry, material behavior, and light arriving after interactions elsewhere in the scene. A pale wall beside a red object may receive reddish reflected light even when the main lamp is white.

Light transport provides a framework for accounting for those relationships. Direct illumination reaches a surface from a light source without an intervening surface reflection. Indirect illumination arrives after other interactions, producing effects such as color bleeding and complex reflected lighting.

Many rendering systems use geometric optics and radiometry without simulating every wave-optical effect. Their assumptions should match the intended image or measurement.

How it works

Specify the scene geometry, materials, light sources, camera, and any participating media such as fog. A surface light-transport equation relates outgoing radiance to emitted radiance plus incoming radiance scattered toward the outgoing direction.

The scattering term combines illumination from many directions with the material's directional response and geometric factors. Incoming illumination depends on other parts of the scene, making the calculation recursive.

Rendering algorithms approximate this relationship in different ways. Path tracing samples sequences of interactions and averages their contributions. More samples can reduce sampling noise, but an incorrect material or light model remains incorrect regardless of sample count.

Volume transport extends the description to absorption and scattering between surfaces. Exposure, tone mapping, and display encoding then affect how the computed light is shown; those presentation stages should not be confused with the underlying transport itself.

Three practical examples

1. Product visualization: A renderer predicts how a glossy object reflects its surroundings so a design team can compare surface finishes under controlled lighting.

2. Synthetic training data: A vision team varies lights and materials in rendered scenes to study whether a detector remains reliable across different appearances.

3. Scene reconstruction: An inverse-rendering system estimates illumination or material parameters by comparing rendered predictions with photographs, while accounting for ambiguity among possible explanations.

Think of it this way

Think of sound traveling through a room, reflecting from walls and being absorbed by furnishings before reaching a microphone. The received signal depends on the whole environment, not only the source. The analogy is limited because light and sound have different physical behavior and scales, and a light-transport model uses optical quantities and material responses.

Worked example

Consider a simplified energy-accounting example. A surface receives 100 units of incident light energy over a specified interval. Assume it reflects 60%, transmits 10%, and absorbs 30%, with no emission or other energy conversion in this example.

The reflected amount is 60 units, the transmitted amount is 10 units, and the absorbed amount is 30 units. These sum to the original 100 units, satisfying the assumed energy balance.

Suppose geometry allows one quarter of the reflected energy to reach a second surface. That surface receives 15 units from this interaction. If it reflects 40% of that incoming amount, it sends 6 units onward in total, distributed according to its directional scattering behavior.

This accounting does not say that a camera receives all 6 units. Its position, viewing direction, aperture, visibility, and the material response determine the fraction reaching it. Nor are these numbers pixel values. They illustrate successive transport and loss; a full image calculation requires radiometric quantities, geometry, and directional integration.

Illustrative case study

Fictional case study: A packaging studio renders a white container beside a strongly colored backdrop. The first preview treats each object with only a simple direct-light approximation, making the container look unnaturally isolated from its environment.

The team adds a transport model that includes indirect illumination and observes a colored contribution on the container. It then compares the result with a controlled reference photograph and adjusts the material and lighting assumptions.

Increasing the rendering samples removes grain but does not fix an initially incorrect gloss setting. The team learns to distinguish numerical noise from modeling error and display choices. The case illustrates diagnostic reasoning rather than a reported gain in rendering accuracy or production speed.

Common misconceptions and mistakes

Light transport is not synonymous with ray tracing. Ray tracing finds geometric intersections, while transport algorithms decide how light contributions are evaluated along possible paths.

More realistic-looking images do not necessarily prove that every material parameter is physically correct. Several parameter combinations can produce similar appearances from one view.

Doubling a light's intensity can double linear radiometric contributions under an unchanged linear model, but it does not guarantee that displayed pixel codes double after tone mapping or clipping.

Sampling noise and model bias are different. Taking more samples can reduce the former without correcting wrong geometry, missing light sources, or unsuitable material assumptions.

Surface-only models do not fully describe scenes where fog, smoke, or another participating medium significantly changes the light between surfaces.

Business use

Transport models support product imagery, architectural visualization, simulation, and synthetic data generation. They can reduce guesswork when teams need to understand how lighting and materials interact, but their usefulness depends on validated inputs and appropriate physical assumptions.

Evaluate representative scenes with known lighting and materials. Compare relevant visual or measured quantities, not just whether the render looks attractive. For synthetic training data, test the downstream model on real captures because physically motivated rendering alone does not guarantee that the simulated distribution covers real-world variation.

People also ask

Does light transport include shadows?

Yes. Visibility determines which light paths are blocked and therefore which contributions can reach a surface or sensor.

Is indirect light always reflected from one surface only?

No. It can involve multiple interactions, each affected by materials, geometry, and absorption.

Can one photograph uniquely reveal all lighting and materials?

Generally no. Different scenes or parameter combinations can produce similar images, creating an inverse problem.

Check your understanding

A non-emitting surface receives 50 energy units and reflects 70% while absorbing the rest. How much is absorbed?

Answer: 15 units. The reflected 35 units and absorbed 15 units sum to the incoming 50 under the stated simplified model.

Related terms

• Rendering equation
• radiance
• global illumination
• path tracing
• BRDF
• participating medium
• inverse rendering

Learn next

Learn radiance and directional material response, then trace a small set of light paths through a simple scene. Separate changes caused by illumination, material parameters, sampling noise, and the display transform.

More in Computer vision and multimodal AI

• Level set - A level set is the collection of points at which a function has the same specified value.
• Levels - Levels is an image-editing adjustment that remaps tonal values using controls for the input black point, input white point, midtones, and output range.
• Light field - A light field describes light traveling through space as a function of position and direction.
• Line intersection - Line intersection is the geometric relationship in which two lines share one or more points.
• Linear filter - A linear filter transforms a signal or image while preserving superposition: filtering a weighted sum of inputs gives the same result as taking the...
• Linear interpolation - Linear interpolation estimates a value between two known samples by following the straight line that connects them.

Sources

Physically Based Rendering, fourth edition, The Light Transport Equation: https://pbr-book.org/4ed/Light_Transport_I_Surface_Reflection/The_Light_Transport_Equation

The primary technical text was opened and its energy-balance and recursive transport explanation checked. Energy accounting and the packaging-studio scenario are original teaching examples.

Previous: Light field Next: Light verbs