With recent advances in camera manufacturing, light field (LF) imaging technology becomes increasingly popular and is commonly used in various applications such as mobile phones, biological microscope, VR/AR etc. Since both intensity and directions of light rays are recorded by LF cameras, the resolution of LF images can be enhanced by using these additional angular information. LF image super-resolution (SR), also known as LF spatial SR, aims at reconstructing high-resolution (HR) LF images from their low-resolution (LR) conterparts. In this repository, we present a collection of papers and datasets on LF image SR, together with their codes and repos.
A more comprehensive overview of LF datasets can be viewed through this link. Here, we only list several popular datasets which are commonly used for LF image SR:
- New light field image dataset (EPFL). website; paper
- Datasets and Benchmarks for Densely Sampled 4D Light Fields (HCIold). paper
- 4D Light Field Dataset (HCInew). website; paper
- INRIA Light field dataset (INRIA). website; paper
- The (New) Stanford Light Field Archive (STFgantry). website
- Stanford Lytro Light Field Archive (STFLytro). website
Model | Published | Codes | Keywords | Better Performance than |
---|---|---|---|---|
Bishop et al. | TPAMI2012 | -- | pioneering work, variational Bayesian, Lambertian prior | -- |
GMM | CVPRW2012 | -- | pioneering work, Gaussian mixture model, subspace projection | -- |
Wanner et al. | ECCV2012 & TPAMI2013 | -- | pioneering work, structure tensor, variational framework | -- |
LFCNN | ICCVW2015 & SPL2017 | TensorFlow | first CNN-based method | GMM, Wanner et al. |
PCA_rr | JSTSP2017 | Matlab | PCA, ridge regression, patch matching | LFCNN, SRCNN, ANR, NCSR |
LFSR_Gul | TIP2018 | -- | CNN | LFCNN, DRRN |
GB | MMSP2017 & TIP2018 | Matlab | graph-based regularization | Wanner et al., GMM, SRCNN |
LFNet | TIP2018 | Theano | bidirectional recurrent CNN | LFCNN, FSRCNN, VDSR, BRCN, DRRN |
LRP | TPAMI2018 | -- | low-rank prior, SVD, QR decomposition, CNN | LFCNN, PCA_rr, VDSR |
LFBM5D | ICIP2018 | Matlab/C++ | entend SR-BM3D to LFs | PCA_rr, SR-BM3D |
LF-DCNN | SPL2018 | -- | EDSR + EPI enhancement | LFCNN, GB, EDSR |
LFSSR | TIP2018 | MatConvNet | spatial-angular separable convolution | LFCNN, PCA_rr, GB, VDSR, LapSRN |
MCSTF | IJCV2019 | -- | motion-compensated spatio-temporal filtering | LFBM5D, GB, SRCNN, VSR, DeepSR, SPMC |
EPGB | TMM2019 | -- | edge-preserved graph-based regularization | GMM, LFNet, GB, SRCNN |
LF-FusNet | ICIP2017 & TCSVT2019 | Caffe | internal and external similarity | PCA_rr, LFNet, GB, VDSR |
resLF | CVPR2019 | PyTorch | multi-branch inputs, residual blocks | GMM, Wanner et al., LFCNN, LFNet, GB, EDSR |
Zhu et al. | CVPRW2019 | TensorFlow | epipolar volume autoencoder, adversarial loss | GB, SRGAN |
HDDRNet | TPAMI2019 & AAAI2020 | TensorFlow | 4D convolution, dense residual connection | LFCNN, PCA_rr, LFNet, resLF, VDSR, MSLapSRN, RDN, ESPCN, DUF |
Jin et al. | CVPR2020 | -- | combinatorial geometry embedding, structural consistency regularization | PCA_rr, LFNet, GB, resLF, EDSR |
LF-InterNet | arXiv2020 | PyTorch | spatial-angular interaction | LFBM5D, GB, LFNet, LFSSR, resLF, VDSR, EDSR, RCAN, SAN, SRGAN, ESRGAN |
We would like to thank Zhen Cheng for the helpful discussion and insightful advice regarding this work.