A 2026 dataset from researchers at Sun Yat-sen University and Nanyang Technological University. It contains 2.5 million degraded images generated from 10,000 original images across three vision tasks: Image Classification, Object Detection, and Instance Segmentation. The dataset was created by applying 10 distortion types across 5 levels and 3 region patterns, with quality scores generated by 75 models.
Use Cases
- Training and benchmarking IQA models based on the 10 distortion types and 5 severity levels.
- Studying the impact of region-specific distortions on vision tasks based on the 3 region patterns.
- Evaluating the correlation between automated quality scores from 75 models and task performance on ImageNet and COCO datasets.
Strengths
- Large scale with 2.5 million images derived from 10,000 originals.
- Systematic degradation coverage with 10 distortion types, 5 levels, and 3 region patterns.
- Grounding in established benchmarks, using images from ImageNet and COCO for three core vision tasks.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Freshness should be verified as the last update date is in the future (2026-04 30).
Provenance
- Source
- Sun Yat-sen University and Nanyang Technological University
- Collection Method
- Images were synthetically degraded from 10,000 originals taken from ImageNet and COCO datasets.
- Time Range
- null
- Freshness
- Last updated 2026-04-30 14:38:19
- Geography
- null