USTC-TD: USTC Test Dataset

USTC-TD is proposed for testing and evaluating the practical image/video coding algorithms. USTC-TD contains 10 video sequences and 40 images with a variety of content coverage. For the image dataset, each image is captured with a high spatial resolution (4K) in RGB, YUV444 color space, and PNG file format. For the video dataset, each video sequence consists of 96 frames, and each frame is captured at 30 fps with 1080P spatial resolution in RGB, YUV444, YUV420 color space, and PNG/YUV file format. For the construction of datasets, we collected the data with the design of different content elements (motion, texture, illumination, view, et al), which aims to cover as close as possible to the real-world coding transmission scenes. Compared with the previous common test image/video datasets, we use different quantitative criteria to comprehensively evaluate the diversity of the proposed dataset from the perspective of spatial, temporal, colorfulness, and lightness information, and demonstrate the wide coverage for various image/video features.