Light100K: Low-Light Image Dataset with Paired Enhancement Targets

Light100K is a dataset of low-light images paired with five enhancement targets. The dataset is organized as Parquet shards with one aligned sample per row, containing the low-light input and five target images as Hugging Face image columns. It was created by ControlLight and last updated on May 26, 2026.

Use Cases

• Training low-light image enhancement models based on the paired low-light input and five target images.
• Benchmarking image restoration algorithms based on the aligned control and target columns.
• Developing multi-target image processing pipelines based on the five enhancement targets (target_l01 ... target_l05).

Strengths

• Contains five enhancement targets per low-light input, providing multiple reference outputs.
• Keeps original PNG bytes with no image decode/re-encode during export, preserving image quality.
• Organized as Parquet shards for better sample-level random access compared to tar-per-folder packaging.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

ControlLight via Hugging Face

Freshness

Last updated 2026-05-26 04:29:20; freshness should be verified.