Techart Conference

Minseob Park 

■ Abstract

Background and Motivation

In modern digital content production, real-time visualization and previsualization are becoming increasingly important. Creators often face difficulties when reconstructing 3D scenes on-site due to the complexity of equipment, data acquisition processes, and the need for precise camera calibration. Existing techniques such as Neural Radiance Fields (NeRF) offer high-quality scene synthesis but are limited by their high computational cost and slow rendering speeds. Therefore, there is a growing demand for a lightweight and efficient solution that enables immediate feedback during creative workflows.

Project Objective

This project aims to develop an AI-based previsualization solution that allows creators to rapidly reconstruct 3D scenes in real-time using minimal input data. By integrating 3D Gaussian Splatting (3DGS) and other complementary technologies, the system enables intuitive and efficient production planning directly at the filming site. The ultimate goal is to provide a tool that simplifies technical barriers and enhances the flexibility of content creation workflows.

Technologies and Methodology

The core rendering technology adopted in this project is 3D Gaussian Splatting, which synthesizes continuous 3D scenes from a set of discrete images. Unlike NeRF, which relies on neural sampling and volumetric rendering, 3DGS represents each point in the scene as a 3D Gaussian and renders the scene via alpha blending. This allows for significantly faster rendering without compromising visual quality, making it ideal for live production settings. To address the challenge of camera pose estimation without prior calibration, the project integrates DUSt3R, a pre-trained Transformer-based method that automatically infers pointmaps and aligns multi-view images. This reduces the complexity of data preparation and allows creators to work with freely captured image sets. Furthermore, in dynamic shooting environments, SLAM3R is used to reconstruct temporally consistent 3D scenes from RGB video input. The incorporation of Fast3R enables efficient parallel  processing, further enhancing real-time performance. In addition to spatial reconstruction, the project explores avatar generation technologies such as Animatable Gaussians and GoM Avatar. These techniques enable the creation of actor-like 3D avatars from minimal image input. Combined with pose estimation using OpenPose, the system supports intuitive previsualization of actor performances, including motion simulation from various camera angles.

Expected Outcomes and Contributions

The proposed solution dramatically reduces the barrier to entry for 3D scene visualization, enabling creators to quickly generate and interact with virtual environments without extensive technical setup. It provides a lightweight, scalable, and AI-driven toolset suitable for real-time previsualization in film, virtual production, and immersive content development. By combining the speed of 3DGS with intelligent automation via DUSt3R and SLAM3R, the system enhances productivity, minimizes latency in decision-making, and empowers creators with a more fluid and creative production experience. Furthermore, by incorporating avatar simulation capabilities, the solution opens new possibilities for visual experimentation and actor blocking prior to full-scale production.

■ Bio

Currently pursuing a Master’s degree in Film at Chung-Ang University’s Graduate School of Advanced Imaging. After directing two short films, Journey (2018) and Rumor (2019), he worked as a cinematographer and production PD in commercial film, advertising, and production settings. Building on extensive experience in large-scale drama and film projects, he served as the cinematographer for the independent feature Mime and entered graduate studies in 2024 to further refine his craft. At the university, he continues his work as a cinematography and lighting director on various short films. Drawing on his background in numerous short and commercial productions, he focuses on creating well-crafted films that combine innovative visual aesthetics and compelling narratives.