Reproduce “USTC-TD: A Test Dataset and Benchmark for Image and Video Coding in 2020s” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhuoyuan Li, Junqi Liao, Chuanbo Tang, Haotian Zhang, Yuqi Li, Yifan Bian, Xihua Sheng, Xinmin Feng and 5 more
Researchers at the University of Science and Technology of China developed USTC-TD, a novel image and video coding test dataset with enhanced diversity in spatial, temporal, and imaging factors. Benchmarking with USTC-TD reveals advanced learned image codecs achieve higher perceptual quality, while traditional video codecs exhibit superior robustness in complex motion scenarios under objective quality metrics.
Key insights
- USTC-TD's diverse content, particularly its emphasis on complex spatial textures, extreme illumination, and varied temporal dynamics, provides a more rigorous and diagnostic testbed for compression algorithms than previous datasets.
- While learned image compression schemes generally achieve superior perceptual quality and objective performance in many scenarios, their generalization can be limited in highly atypical or challenging conditions compared to traditional codecs.
- Traditional video codecs maintain a robust advantage over learned codecs in handling complex motion and diverse temporal correlations under objective metrics (PSNR, VMAF), highlighting persistent challenges for learned video compression in motion modeling and error propagation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.