---
title: "Bidirectional Estimators for Light Transport"
authors:
  - "Eric Veach"
  - "Leonidas J. Guibas"
year: 1994
venue: "EGWR"
arxiv: null
doi: null
source: "https://graphics.stanford.edu/papers/bidir/"
topics:
  - gi
  - bdpt
seed_rank: 376
seed_batch: "systems-prefill-2026-08-13"
reviewed: "2026-08-13"
pool: "graphics"
relevance_score: 8
cites:
  - title: "The rendering equation"
    url: "https://doi.org/10.1145/15886.15902"
    year: 1986
    arxiv: null
    doi: "10.1145/15886.15902"
  - title: "Distributed ray tracing"
    url: "https://doi.org/10.1145/800031.808590"
    year: 1984
    arxiv: null
    doi: "10.1145/800031.808590"
  - title: "Adaptive radiosity textures for bidirectional ray tracing"
    url: "https://doi.org/10.1145/97879.97895"
    year: 1990
    arxiv: null
    doi: "10.1145/97879.97895"
see:
  - "152-the-rendering-equation"
---

# Bidirectional Estimators for Light Transport

## One-sentence takeaway

Every light-to-eye path can be generated by several partitions (k vertices from the light, the rest from the eye), and those partitions are different unbiased estimators with wildly different variance.

## Why it matters here

This is the paper that invents bidirectional path tracing as a family of estimators, before MIS in the 1995/1997 work. Anoptic’s offline reference path and any future BDPT/VCM debug view start here: connect a light subpath to an eye subpath instead of hoping a camera path hits a lamp.

## Key ideas

- A path of length s+t can be sampled by tracing s vertices from a light and t from the eye, then connecting the tips; each (s,t) is unbiased for the same measurement.
- Some families are cheap for caustics (light-first), others for view-dependent glossy (eye-first); no single partition wins everywhere.
- Combining the estimators (the “work in progress” in this paper) becomes multiple importance sampling in Veach’s later thesis.
- Avoids meshing: the pitch is Monte Carlo vs radiosity for general geometry and BRDFs.

## Caveats

## Links

- Stanford page: https://graphics.stanford.edu/papers/bidir/
- PostScript (full): https://graphics.stanford.edu/papers/bidir/bidir-bw.ps
- Thesis chapter: https://graphics.stanford.edu/courses/cs348b-03/papers/veach-chapter10.pdf