Fast and Efficient Restoration of Extremely Dark Light Fields

Abstract

restoring-extremely-dark-light-fields-computational-imaging-lab-iit-madras

Key Highlights

Description

Results

restoring-extremely-dark-light-fields-main-result-computational-imaging-lab-iit-madras

Figure 4: Visual and epipolar comparison of restorations achieved by our method vs. existing approaches on the L3F-100 dataset.

restoring-extremely-dark-light-fields-quantitative-table-psnr-ssim-fps-computational-imaging-lab-iit-madras

Figure 1. Inference speed vs. restoration quality ➯ our method is both faster and higher quality than prior approaches.

restoring-extremely-dark-light-fields-input-computational-imaging-lab-iit-madras

Figure 5. More visual and epipolar comparisons vs. the state-of-the-art L3Fnet ➯ our method restores finer details.

restoring-extremely-dark-light-fields-baseline-computational-imaging-lab-iit-madras

Figure 6. Generalization capability of our network in restoring light fields beyond the training distribution.

Visualizations

restoring-extremely-dark-light-fields-pipeline-animation-computational-imaging-lab-iit-madras

Animation: Step-by-step walkthrough of the restoration pipeline, from raw noisy light field input to the final restored output.

restoring-extremely-dark-light-fields-pipeline-architecture-computational-imaging-lab-iit-madras

Pipeline Architecture. Stage-wise breakdown of the processing pipeline used to go from input to output.

restoring-extremely-dark-light-fields-radiance-intervals-computational-imaging-lab-iit-madras

Radiance / Intensity Intervals. Illustration of the intensity ranges handled across the processing stages.

restoring-extremely-dark-light-fields-project-overview-computational-imaging-lab-iit-madras

Project Overview. High-level summary of the problem setting, approach, and key results.

Publications