講演情報

[8a-N205-6]Implicit Neural Representation of Light Fields via Decomposed Spatial Geometry and Color

〇Jun Qiu1, Ligen Shi1, Chang Liu1 (1.Beijing Information Science and Technology Univ.)

キーワード:

Computational Imaging、Light field、Implicit Neural Representation

This work proposes a novel Implicit Neural Representation for light-field imaging, addressing the spatial-angular resolution trade-off by decomposing the LF into geometry and color fields. A two-stage training strategy is employed to separately learn spatial offsets and central-view radiance, enabling efficient and interpretable LF reconstruction. Experiments show that the method achieves high-quality LF super-resolution with fewer parameters and no need for large-scale dataset.