Knives And Time: Damaged Images with Semi-Automated Annotation

A custom collection of paintings, images, and photographs exhibiting various types of damage. The dataset was created via manual collection and semi-automated annotation, with an initial sweep using the BLIP model followed by manual refinement. It was last updated on May 5, 2024, by the author 'calm-and-collected'.

Use Cases

• Train image restoration models based on examples of abrasion, water damage, and scratches.
• Develop classifiers for damage types based on annotated categories like 'torn', 'burned', 'cut', and 'pierced'.
• Benchmark model robustness based on images with over composure or other artifacts.
• Study the effects of physical degradation on visual content based on the described collection method.

Strengths

• Focuses on a specific niche of damaged visual media, which is less common than pristine image datasets.
• Annotation process combines automated (BLIP) and manual methods, suggesting a balance of scale and precision.
• Source material is from the public domain or uses a CC0 license, implying fewer legal restrictions.

Limitations

• Description metadata is limited; actual data quality, size, and column structure require manual inspection after download.
• Row count, file formats, and specific license details are unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific sources and collection methods used by the author.

Provenance

Source

huggingface, author calm-and-collected

Collection Method

Manually collected from public domain or CC0 sources, with semi-automated annotation using BLIP followed by manual work.

Time Range

null

Freshness

Last updated 2024-05-05 11:13:22; freshness should be verified.

Geography

null