mkkim007 / Bokeh-Rendering-with-Vision-Transformers

Establishing new state-of-the-art results for Bokeh Rendering on the EBB! Dataset.

Bokeh-Rendering-with-Vision-Transformers

Establishing new state-of-the-art results for Bokeh Rendering on the EBB! Dataset. The preprint of our work can be found here.

Sample:

References

Model adapted from https://github.com/isl-org/DPT

SSIM loss can be found at https://github.com/Po-Hsun-Su/pytorch-ssim

MSSSIM loss can be found at https://github.com/jorge-pessoa/pytorch-msssim

LPIPS can be found at https://github.com/richzhang/PerceptualSimilarity

BRViT Weights

Our latest model weights can be downloaded from here.

BRViT Metrics

Common Metrics with the latest weights for model comparison:

1. PSNR: 24.76
2. SSIM: 0.8904
3. LPIPS: 0.1924