A5087784761- Advanced Image Processing Techniques
- Advanced Vision and Imaging
- Image Processing Techniques and Applications
- Image and Signal Denoising Methods
- Image Enhancement Techniques
- Image and Video Quality Assessment
- Advanced Image Fusion Techniques
- Digital Media Forensic Detection
ETH Zurich
2019-2023
With the recent trend for ultra high definition displays, demand quality and efficient video super-resolution (VSR) has become more important than ever. Previous methods adopt complex motion compensation strategies to exploit temporal information when estimating missing frequency details. However, as estimation problem is a highly challenging problem, inaccurate may affect performance of VSR algorithms. Furthermore, module also introduce heavy computational burden, which limits application...
Many super-resolution (SR) models are optimized for high performance only and therefore lack efficiency due to large model complexity. As often not practical in real-world applications, we investigate propose novel loss functions, enable SR with perceptual quality from much more efficient models. The representative power a given low-complexity generator network can be fully leveraged by strong guidance towards the optimal set of parameters. We show that it is possible improve recently...
Video super-resolution (VSR) has many applications that pose strict causal, real-time, and latency constraints, including video streaming TV. We address the VSR problem under these settings, which poses additional important challenges since information from future frames is unavailable. Importantly, designing efficient, yet effective frame alignment fusion modules remain central problems. In this work, we propose a recurrent architecture based on deformable attention pyramid (DAP). Our DAP...
This paper reviews the extreme video super-resolution challenge from AIM 2019 workshop, with emphasis on submitted solutions and results. Video x16 is a highly challenging problem, because 256 pixels need to be estimated for each single pixel in low-resolution (LR) input. Contrary image (SISR), provides temporal information, which can additionally leveraged restore heavily downscaled videos imperative any (VSR) method. The composed of two tracks, find best performing method fully supervised...
This paper reviews the NTIRE 2020 challenge on video quality mapping (VQM), which addresses issues of from source domain to target domain. The includes both a supervised track (track 1) and weakly-supervised 2) for two benchmark datasets. In particular, 1 offers new Internet benchmark, requiring algorithms learn map more compressed videos less in training manner. 2, are required one device another when their varies substantially weakly- aligned pairs available. For 1, total 7 teams competed...
The current rapid advancements of computational hardware has opened the door for deep networks to be applied real-time video processing, even on consumer devices. Appealing tasks include super-resolution, compression artifact removal, and quality enhancement. These problems require high-quality datasets that can training benchmarking. In this work, we therefore introduce two datasets, aimed a variety tasks. First, propose Vid3oC dataset, containing 82 simultaneous recordings 3 camera...
Abstract Video enhancement is a challenging problem, more than that of stills, mainly due to high computational cost, larger data volumes and the difficulty achieving consistency in spatio-temporal domain. In practice, these challenges are often coupled with lack example pairs, which inhibits application supervised learning strategies. To address challenges, we propose an efficient adversarial video framework learns directly from unpaired examples. particular, our introduces new recurrent...
This paper reviews the video extreme super-resolution challenge associated with AIM 2020 workshop at ECCV 2020. Common scaling factors for learned (VSR) do not go beyond factor 4. Missing information can be restored well in this region, especially HR videos, where high-frequency content mostly consists of texture details. The task is to upscale videos an 16, which results more serious degradations that also affect structural integrity videos. A single pixel low-resolution (LR) domain...
Video enhancement is a challenging problem, more than that of stills, mainly due to high computational cost, larger data volumes and the difficulty achieving consistency in spatio-temporal domain. In practice, these challenges are often coupled with lack example pairs, which inhibits application supervised learning strategies. To address challenges, we propose an efficient adversarial video framework learns directly from unpaired examples. particular, our introduces new recurrent cells...
With the recent trend for ultra high definition displays, demand quality and efficient video super-resolution (VSR) has become more important than ever. Previous methods adopt complex motion compensation strategies to exploit temporal information when estimating missing frequency details. However, as estimation problem is a highly challenging problem, inaccurate may affect performance of VSR algorithms. Furthermore, module also introduce heavy computational burden, which limits application...
This paper reviews the NTIRE 2020 challenge on video quality mapping (VQM), which addresses issues of from source domain to target domain. The includes both a supervised track (track 1) and weakly-supervised 2) for two benchmark datasets. In particular, 1 offers new Internet benchmark, requiring algorithms learn map more compressed videos less in training manner. 2, are required one device another when their varies substantially weakly-aligned pairs available. For 1, total 7 teams competed...
Video super-resolution (VSR) has many applications that pose strict causal, real-time, and latency constraints, including video streaming TV. We address the VSR problem under these settings, which poses additional important challenges since information from future frames is unavailable. Importantly, designing efficient, yet effective frame alignment fusion modules remain central problems. In this work, we propose a recurrent architecture based on deformable attention pyramid (DAP). Our DAP...
Many super-resolution (SR) models are optimized for high performance only and therefore lack efficiency due to large model complexity. As often not practical in real-world applications, we investigate propose novel loss functions, enable SR with perceptual quality from much more efficient models. The representative power a given low-complexity generator network can be fully leveraged by strong guidance towards the optimal set of parameters. We show that it is possible improve recently...