[LS2N_IPI_SR] Does Super-Resolution Survive Compression? A Subjective and Objective Study of Image vs. Video Quality for Modern SR
Abstract
Super-resolution (SR) is increasingly used to enhance user-generated, legacy and AI-generated video before distribution, yet how much quality it delivers once the content is compressed remains unclear. A second, practical question follows: which metric should serve as the objective when enhancement and compression are optimised jointly? This paper presents a large-scale subjective study of SR within a complete delivery chain: 30 source contents, four generations of SR algorithms, three delivery resolutions and four quality levels, giving 900 processed sequences scored twice, by 11 718 pairwise comparisons on still frames and 26 068 ACR ratings on video. Against a conventional 540p chain, SR does not reduce the bitrate required to reach a given quality, but it reaches a quality that chain never attains on the tested ladder, by a median +0.35 MOS for Real-ESRGAN and on 90% of the contents. The two experiments disagree systematically: gains that are clear on still frames largely vanish on video, which quantifies the cost of temporal inconsistency and is most pronounced for the diffusion-based method. Benchmarking 21 objective metrics on the same stimuli shows that none predicts these scores reliably (SRCC up to 0.600). They order compression levels well (up to 94% correct), but at a fixed rate the SR models can be ranked as low as 66% correct by PSNR, while learned full-reference networks reach 91%. The dataset is made publicly available to support future research.
Read the paper
Related studies
- Learning from First-Generation Qualitative Approaches in the IS Discipline: An Evolutionary View and Some Implications for Authors and Evaluators (PART 1/2)
- Commentary on the clinical and preclinical dosage limits of interstitially administered magnetic fluids for therapeutic hyperthermia based on current practice and efficacy models
- Prosthesis: New Technological Opportunities and Innovative Biomedical Devices
- The history and future of neural modeling for cochlear implants
- Ryanodine receptors: waking up from refractoriness
- Novel Transgenic Rabbit Model Sheds Light on the Puzzling Role of Matrix Metalloproteinase-12 in Atherosclerosis
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.