FS-LSPD: Few-Shot Multi-Label Dataset for Sensitive Image Classification
Description
FS-LSPD is a few-shot multi-label image classification dataset for sensitive-content recognition. It provides image-level multi-label annotations for four target categories: breast, male_genital, female_genital, and anus. Each image is represented by a four-dimensional binary label vector, allowing multiple target categories to be assigned to the same image. Images without any target category are treated as non-sensitive samples and assigned the all-zero label vector. The dataset includes controlled few-shot training splits for 1-shot, 5-shot, 10-shot, 20-shot, and 30-shot settings. Each shot setting contains five subsets to support evaluation across different sampled support sets. The few-shot splits follow fixed label-composition rules so that every target label appears exactly K times in each K-shot setting. Non-sensitive person images are also included as all-zero negative samples. The public repository provides the prepared CSV annotation files, few-shot split files, a fixed test CSV file, and baseline training/evaluation code. The CSV files include image metadata, four binary label columns, and the number of positive labels. Face regions in sensitive images were anonymized by masking before data release to protect personal identity. Due to privacy, copyright, and redistribution restrictions associated with sensitive visual content, the image files are not publicly redistributed. Researchers who wish to access the anonymized training and test images must complete the dataset access request form and agree to the dataset usage terms. Access is provided for research purposes only after approval. For further information or access-related questions, please contact the corresponding author at lungvd@uit.edu.vn.
Files
Steps to reproduce
To reproduce the dataset construction and baseline experiments, please follow the instructions provided in the GitHub repository: https://github.com/chequanghuy/FS-LSPD.