Sign in to view source links and access this dataset
Description
4KLSDB is a dataset of 133,468 native-4K images for computer vision tasks. It contains 129,484 training, 2,000 validation, and 1,984 test images across categories like nature, urban scenes, people, and artwork. The dataset was created by SingleBicycle and presented at CVPR 2026.
Use Cases
Train super-resolution models based on the dataset's native-4K image collection.
Develop text-to-image generation models using the described 4K image-text pairs.
Benchmark image restoration algorithms on the provided validation and test splits.
Conduct research on high-fidelity image synthesis across diverse categories like urban scenes and artwork.
Strengths
Contains 133,468 total images, providing a substantial scale for model training.
Offers a dedicated split of 1,984 test images for reliable evaluation.
Covers a diverse range of eight categories including nature, urban scenes, people, and animals.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Data may reflect bias inherent to the collection sources and methods on Hugging Face.
Provenance
Source
SingleBicycle on Hugging Face.
Freshness
Last updated 2026-05-26 05:18:27; freshness should be verified.
License is unknown; terms of use must be verified before application.