Bluesky Moderation Image Collection With 69,247 Examples

69,247 images categorized into classes including normal, sexual, violence, and weapon, collected for the Mod-RWKV Bluesky Hackathon project. The dataset totals 10.57 GB and is intended for image classification tasks in the NSFW domain.

Use Cases

• Train an image classifier to distinguish between the 'normal', 'sexual', 'violence', and 'weapon' categories using the 69,247 labeled examples.
• Fine-tune a vision model for NSFW content detection by leveraging the explicit 'sexual' and 'violence' class labels.
• Analyze the distribution and visual features of images within the 'weapon' category for content moderation benchmarking.
• Build a multi-label or hierarchical classifier if images contain overlapping concepts from the provided categorical labels.

Strengths

• Contains 69,247 image examples, providing a substantial sample for model training.
• Explicitly labeled for four distinct moderation-relevant categories: normal, sexual, violence, and weapon.
• Dataset size of 10.57 GB indicates high-resolution or numerous image files.

Limitations

• Class distribution is unknown, posing a risk of severe imbalance which could bias classifiers.
• Label quality and consistency are not verified; potential noise exists from the original collection method.
• Limited to four broad categories, which may not capture nuanced sub-types of harmful content.

Provenance

Source

Hugging Face dataset by author umuthopeyildirim.

Collection Method

Collected for the Mod-RWKV Bluesky Hackathon project.

Time Range

null

Freshness

Last updated on 2024-03-01.

Geography

Region tag suggests 'us', but specific spatial coverage is unknown.