PNV-ObsceneDB1K

The proliferation of online content has raised significant concerns regarding the identification and regulation of obscene content, necessitating robust datasets for research and development. This study presents a curated dataset comprising 1,000 videos, equally divided into 500 obscene and 500 non-obscene videos. These were collected from various internet and online platforms. These videos are from various languages and regions (e.g., Hindi, English, Japanese, and other regional languages) and content types (e.g., adult scene, nudity, suggestive). Each video label is an obscene or non-obscene. To ensure labeling quality, a multi-annotator approach was employed to address the subjective nature of obscene content, with subcategories defined for precise classification. The videos range in duration from 10 seconds to 17 minutes, offering a diverse and representative sample for analyzing obscene content. This dataset serves as a valuable resource for developing AI models content moderation systems and aimed at mitigating harmful online content.