Crack is a collection of annotated images for road crack detection, aggregated by Mr-Perfectuz on HuggingFace. It comprises three subsets: BJN260_full_resolution with 260 night scene images from Beijing, Rain365_full_resolution with 365 post-rain daytime images, and Sun520_full_resolution with 520 sunny daytime images. Each subset provides crack photos, pixel-level ground-truth masks, and composite overlay images.
Use Cases
- Training semantic segmentation models based on provided pixel-level ground-truth masks.
- Benchmarking crack detection algorithms across different lighting and weather conditions mentioned in the description.
- Developing automated road inspection systems based on annotated crack imagery from varied scenarios.
Strengths
- Provides pixel-level ground-truth masks for precise model training and evaluation.
- Contains 1,145 total images across three distinct environmental scenarios: night scenes (260), post-rain daytime (365), and sunny daytime (520).
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download.
- Column-level documentation is absent; field semantics must be inferred after download.
Provenance
- Source
- Mr-Perfectuz on HuggingFace, citing academic work by Li et al. (2021).
- Freshness
- Last updated 2026-06-15 08:18:09; freshness should be verified.
- Geography
- Includes data from Beijing, China, based on the BJN260 subset description.