Compressed Radiance Fields Coding for Memory-Efficient Representations and Quality Trade-offs

Radiance field representations have emerged as a powerful approach for novel view synthesis and 3D scene reconstruction, but their high memory consumption remains a major limitation for large-scale and real-time applications. This thesis investigates compression and coding strategies for radiance field representations with the goal of achieving memory-efficient models while preserving visual quality. Representative radiance field methods are implemented and extended with compression techniques such as parameter reduction, quantization, and structural simplification. The impact of these techniques is evaluated through a systematic quality and performance assessment, considering visual fidelity, rendering efficiency, and memory footprint. Quantitative metrics and perceptual similarity, together with qualitative visual analysis, are used to characterize the trade-offs between compression level and reconstruction quality. 

In addition, the thesis is conducted under the supervision of the lead chair of the JPEG Radiance Fields initiative, offering the opportunity to align experimental findings with emerging standardization efforts and to explore how research-driven insights can contribute to the development of interoperable and efficient radiance field coding frameworks. The outcomes aim to provide both practical guidelines for memory-efficient radiance field deployment and relevant input for future standardization activities.

Master's degree: Master of Engineering Science

Required educational background: Computer Science, Electrotechnics/Electrical Engineering

For more information or application, please contact the supervising scientist Saeed Mahmoudpour (mahmoudpour.saeed@imec.be).