Instance level detection of thoracic diseases or abnormalities are crucial for automatic diagnosis in chest X-ray images. Most existing works on chest X-rays focus on disease classification and weakly supervised localization. In order to push forward the research on disease classification and localization on chest X-rays. We provide a new benchmark called ChestX-det10, including box-level annotations of 10 categories of disease/abnormality of $\sim$ 3,500 images. The annotations are located at https://github.com/Deepwise-AILab/ChestX-Det10-Dataset.