This repository provides a representative subset of the fish fry counting dataset used in the study entitled: "MSSFNet: A Multi-Scale Semantic Fusion Network for Fish Fry Counting in Complex Aquaculture Environments" The dataset was collected from a commercial aquaculture facility located in Linghu Town, Huzhou City, Zhejiang Province, China. The primary targets include silver carp (Hypophthalmichthys molitrix), grass carp (Ctenopharyngodon idella), and grass carp fry. The sample dataset contains representative images collected under diverse aquaculture conditions, including: High-density fish aggregation Low-density fish distribution Different target scales Water turbidity variations Illumination reflections Partial target occlusion These samples are provided to facilitate independent verification of the experimental data and improve the reproducibility of the proposed counting framework.