AlgaeDet-HR: A high-resolution bright-field microscopic image dataset for multi-class algae detection

AlgaeDet-HR is a high-resolution bright-field microscopic image dataset developed for multi-class algae detection. The dataset contains 1,664 JPEG images at a native resolution of 3072 × 2048 pixels and 30,552 object-level bounding-box annotations covering 13 operational algae categories. The algae were isolated from natural seawater samples collected in 2026 from coastal waters around Pingtan, Ningde, and Quanzhou in Fujian Province, China. The samples were separated, purified, and cultured before imaging. Cultured samples were imaged directly without additional concentration, fixation, or staining. Approximately 20 μL of cultured sample was used for each observation under a consistent 10× bright-field configuration. Image acquisition employed 15-layer sequential scanning followed by software-based image stitching. The released package contains the original microscopic images, source X-AnyLabeling JSON annotations, COCO-format annotations, YOLO-format annotations, Pascal VOC-format annotations, class definitions, image- and instance-level metadata, recommended training/validation/test split files, validation scripts, annotation-conversion scripts, visualization scripts, and baseline experiment utilities. The recommended split contains 1,330 training images, 167 validation images, and 167 test images. Automated quality control verified image readability, image–annotation correspondence, bounding-box validity, annotation-format consistency, and exact duplicate images using MD5 hashes. No exact duplicate image groups were identified in the released dataset. The dataset is intended to support microscopic algae detection, small-object detection, dense-object detection, algae counting, computer vision benchmarking, harmful algal bloom surveillance, and automated water-quality monitoring.