This work presented a new drone-based face detection dataset Drone LAMS in order to solve issues of low performance of drone-based face detection in scenarios such as large angles which was a predominant working condition when a drone flies high. The proposed dataset captured images from 261 videos with over 43k annotations and 4.0k images with pitch or yaw angle in the range of -90° to 90°. Drone LAMS showed significant improvement over currently available drone-based face detection datasets in terms of detection performance, especially with large pitch and yaw angle. Detailed analysis of how key factors, such as duplication rate, annotation method, etc., impact dataset performance was also provided to facilitate further usage of a drone on face detection.