4D Light Field Benchmark

Check out the new Google group to connect with other light field researchers, discuss ideas, and initiate collaborations. Some questions about the benchmark are answered over there as well! There is also a new collection of light field resources. Please feel encouraged to add further links.

go back to table

Acronym

LFNASNet

Method title

Fully Convolutional Neural Network using Neural Architecture Search

Method description

Fully convolutional neural network for disparity estimation from light field sub-aperture images.
- Input: Stack of Grayscale sub-aperture images for four angular directions (0, 45, 90, 135)
- Hidden Layers: Three Hidden Layers in the Multi-stream Network followed by Six Hidden Layers in a Single Stream Network

Camera setup

Cross + X shape (horizontal and vertical 9 views)

Parameter description

Network Parameters: 3,187,969

Programming language

Python using Tensorflow-Keras Library

Runtime environment

Rocky Linux 8.8 (Green Obsidian), Intel(R) Xeon(R) Gold 6338 CPU @ 2GHz, NVIDIA Tesla A100 HGX GPU with 40GB RAM

Additional data

Raw evaluation scores for all scenes