以示意圖片推測三維場景打光參數

Abstract

在這個論文中，我們介紹了一套解決方案以對付『以示意圖片推測三維場景打光參數』的問題： 給定一個場景資訊以及一張打光示意圖，程式可以自動找出能產生最近似示意圖的繪圖結果的打光設定。 我們的系統由四大模組組成：初期推測，最佳化，再打光，圖片誤差計算。 由於本問題是解空間維度非常龐大的非線性最佳化問題，因此好的初期推測決定了結果的好壞。 針對這工作，我們提出了高效率與高成效的演算法。 實驗結果顯示我們的系統可以產生品質良好的打光設定。 自動化的打光設定推測也能夠大大地減輕打光師的負擔。

In this thesis, we introduce and present a solution to the lighting-by-guide problem: given the description of a scene and an image as the lighting guide, automatically find a lighting configuration so that the resulted rendering best matches the lighting guide. Our solution framework consists of four main modules, initial guess, optimization, relighting, and difference metric evaluation. Since the resulted problem is a high-dimensional nonlinear optimization problem, it is crucial to have a good initial guess. We propose an efficient and effective algorithm for this task. Experiments show that our system can produce good quality lighting setups. The result is an automatic lighting algorithm which infers the lighting configuration from the lighting guide, greatly alleviating lighting artists' workload.