5,666 scene photographs annotated for mirror surfaces, introduced at CVPR 2020. The dataset includes 5,095 training images with masks and edges, and 571 test images with masks only. It was created by author 'garrying' and last updated on the Hugging Face platform in April 2026.
Use Cases
- Train mirror detection models based on annotated mask images.
- Benchmark segmentation algorithms on the provided test set.
- Develop edge detection techniques for reflective surfaces.
- Evaluate model performance on a dedicated computer vision challenge.
- Research the properties of mirror surfaces in natural scenes.
Strengths
- 5,666 total images provide a substantial corpus for training and testing.
- 5,095 training images include both mask and edge annotations.
- 571 test images allow for standardized benchmarking.
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
- garrying
- Freshness
- Last updated 2026-04-30 17:20:12; freshness should be verified.