TigerPufferDetect

The released dataset contains 1,600 original JPG images collected from a real aquaculture factory, including 1,000 above-water images and 600 underwater images. Each visible fish instance is manually annotated with a bounding box in XML format and assigned the category label ``fish''.

The above-water images capture dense fish distributions and body overlap near the water surface, while the underwater images include challenging visual conditions such as turbidity, greenish color cast, light attenuation, motion blur, and background interference. This dataset is designed to support research on fish detection, counting, and visual monitoring in smart aquaculture.