NTIRE 2017 Challenge on Single Image Super-Resolution: Methods and Results

By Radu Timofte, Eirikur Agustsson, Luc Van Gool, Ming-Hsuan Yang, Lei Zhang, Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, Kyoung Mu Lee and 66 more. 2017

1,901 citations recorded by the source corpus.

This paper reviews the first challenge on single image super-resolution (restoration of rich details in an low resolution image) with focus on proposed solutions and results. A new DIVerse 2K resolution image dataset (DIV2K) was employed. The challenge had 6 competitions divided into 2 tracks with 3 magnification factors each. Track 1 employed the standard bicubic downscaling setup, while Track 2 had unknown downscaling operators (blur kernel and decimation) but learnable through low and high res train images. Each competition had ∽100 registered participants and 20 teams competed in the final testing phase. They gauge the state-of-the-art in single image super-resolution.

Subfields: Computer Vision and Pattern Recognition, Media Technology, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Media Technology

In Evaluation & Benchmarks

Citations in the corpus

Designing a Practical Degradation Model for Deep Blind Image Super-Resolution (2021, methodology): “Two representative degradation models are bicubic degradation [46] and traditional degradation [28, 45].”
VEGA: Towards an End-to-End Configurable AutoML Pipeline (2020, methodology): “The searched models are trained with dataset DIV2K (Timofte et al., 2017) and and evaluated with Urban 100.”
AutoGAN-Distiller: Searching to Compress Generative Adversarial Networks (2020, positive): “, 2018a) on a combined dataset of DIV2K and Flickr2K (Timofte et al., 2017) with a upscale factor of 4×, following (Wang et al.”
Hierarchical Neural Architecture Search for Single Image Super-Resolution (2020, methodology): “In the experiments, we use the DIV2K dataset [37] to train all the models and conduct comparisons on five benchmark datasets, including Set5 [38], Set14 [39], BSD100 [40], Ur-”
ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks (2018, methodology): “To this end, we further use the Flickr2K dataset [38] consisting of 2650 2K high-resolution images collected on the Flickr website, and the OutdoorSceneTraining (OST) [40] dataset to enrich our training set.”
Enhanced Deep Residual Networks for Single Image Super-Resolution (2017, methodology): “Also, the same model achieves different levels of performance by dif0853 from DIV2K [26] HR (PSNR / SSIM)”
Image Super-Resolution Using Very Deep Residual Channel Attention Networks (2018, methodology): “Following [10, 11, 17, 35], we use 800 training images from DIV2K dataset [35] as training set.”
Towards Real-World Blind Face Restoration with Generative Facial Prior (2021, methodology): “We also employ pyramid restoration guidance [42] for intermediate supervision [62].”
Maintaining Natural Image Statistics with the Contextual Loss (2018, background): “A more comprehensive overview of the current trends can be found in [28].”
Super-Resolution of Sentinel-2 Images: Learning a Globally Applicable Deep Neural Network (2018, background): “, 2017), the state-of-the-art in single-image super-resolution and winner of the NTIRE super-resolution challenge (Timofte et al., 2017).”
Channel-Wise and Spatial Feature Modulation Network for Single Image Super-Resolution (2018, methodology): “We use 800 high-quality training images from DIV2K [48] to train our model.”
Residual Feature Distillation Network for Lightweight Image Super-Resolution (2020, methodology): “DIV2K and Flickr2K datasets are used for training the E-RFDN model.”
NTIRE 2018 Challenge on Single Image Super-Resolution: Methods and Results (2018, methodology): “DIV2K Dataset [1] employed by NTIRE 2017 SR challenge [31] is used also in our challenge.”
Progressive Fusion Video Super-Resolution Network via Exploiting Non-Local Spatio-Temporal Correlations (2019, background): “For SISR, there are some datasets with high quality like BSDS 500 [1] and DIV2K [32].”
Component Divide-and-Conquer for Real-World Image Super-Resolution (2020, methodology): “In the area of SR, widely-used SR datasets include Set5 [1], Set14 [29], BSD300 [17], Urban100 [9] and DIV2K [24].”
Wide Activation for Efficient and Accurate Image Super-Resolution (2018, methodology): “We train our models on DIV2K dataset [35] In this part, we show results of baseline model EDSR [19] and our proposed WDSR-A and WDSR-B for the task of image bicubic x2 super-resolution on DIV2K dataset.”