Species196

Species196是由华为诺亚方舟实验室创建的一个大型半监督数据集，专注于入侵物种的细粒度识别。该数据集包含196个物种类别，总计超过19,236张精确标注的图像（Species196-L）和120万张未标注图像（Species196-U）。Species196不仅覆盖了物种的不同生命周期，还提供了从界到种的详细分类信息，支持多种学习范式，如监督学习、半监督学习、自监督预训练和零样本推理。此数据集对于开发和评估基于深度学习的入侵生物识别系统具有重要价值，特别是在提高模型性能和应对数据稀缺挑战方面。

Species196 is a large-scale semi-supervised dataset created by Huawei Noah's Ark Lab, focusing on fine-grained recognition of invasive species. It encompasses 196 species categories, with a total of over 19,236 accurately annotated images (Species196-L) and 1.2 million unannotated images (Species196-U). Species196 not only covers different life cycles of the target species, but also provides detailed taxonomic information from kingdom to species level, supporting multiple learning paradigms including supervised learning, semi-supervised learning, self-supervised pre-training and zero-shot inference. This dataset holds significant value for developing and evaluating deep learning-based invasive species recognition systems, particularly in improving model performance and addressing the challenge of data scarcity.