Recent advances in conventional and learning-based image coding have increased the demand for benchmark datasets that support fine-grained assessment of compressed image quality, particularly for learning-based image compression methods.
The data in this dataset represent Assessment of Image Coding 2026 (AIC2026), a large-scale dataset for high-fidelity image compression containing 70 source images selected from 2,932 candidates using semantic clustering, inter-metric disagreement among objective image quality assessment (IQA) methods, and manual inspection and refinement. The dataset covers a wide range of compression artifacts produced by 8 conventional and 4 learning-based codecs across 17 coding configurations. Each source image is encoded using 7 codecs. For each source-codec pair, decoded images are provided at 20 perceptually spaced distortion levels, corresponding approximately to 0.2-4.0 JND using CVVDP for distortion estimation, yielding 9,618 distorted images. This fine-grained sampling enables analysis of rate-distortion behavior and objective metric evaluation for subtle quality differences across a wide range of compression artifacts.
An extensive objective analysis using 22 conventional and 12 learning-based IQA methods shows substantial disagreement among current IQA methods for fine-grained quality differences, particularly for artifacts introduced by learning-based codecs.
More information about the dataset including dataset structure, file naming conventions, codec acronyms, accompanying metadata files, citation information, and licensing terms. can be found in the README.
(2026-06-30)
