Chapter 10: Computational photography

• The High Dynamic Range radiance maps captured by Debevec and Malik (1997) at https://www.debevec.org/Research/HDR are still the go-to place to find high-quality HDR images.

• The RealSR real-world super-resolution dataset developed by Cai, Zeng et al. (2019) can be used to train and test SR algorithms on real imaging degradations. This dataset forms the basis for the NTIRE challenges on real image super-resolution (Cai, Gu et al. 2019), which provide empirical comparisons of recent deep network-based algorithms.

• The latest benchmark for comparing image denoising algorithms, the NTIRE 2020 Challenge on Real Image Denoising (Abdelhamed, Afifi et al. 2020), is based on a smartphone image denoising dataset (SIDD) (Abdelhamed, Lin, and Brown 2018) created by averaging sets of real-world noisy images.

• Thea alpha matting evaluation website, http://alphamatting.com (Rhemann, Rother et al. 2009) provides a standard set of test images and a leaderboard.

• The video matting dataset at https://videomatting.com (Erofeev, Gitman et al. 2015) provides stop-motion animation videos created by carefully hand-matting each frame.

• Lin, Ryabtsev et al. (2021) describe a high-resolution real-time video matting system along with two new video and image matting datasets.

• The AIM 2020 Workshop and Challenges on image inpainting (Ntavelis, Romero et al. 2020a) provides datasets for evaluating such algorithms.

C.1 Datasets and benchmarks