The Penn-Fudan pedestrian detection and segmentation dataset was created by researchers from the University of Pennsylvania and Fudan University, and is mainly used for pedestrian detection tasks. This dataset contains 170 high-resolution RGB images, which are captured from video sequences, and there are 0 to 6 pedestrian targets in each image. The position of each pedestrian is accurately marked by a rectangular box (mask), providing bounding box coordinate information for easy target detection training and testing.

The file structure of the dataset is as follows:

• Annotation/: Contains the annotation files for each image.
• PedMasks/: Contains the pedestrian segmentation mask corresponding to each image.
• PNGImages/: Contains all the images in the dataset.

The images were collected from a variety of environments, such as campuses, streets, and crosswalks, covering different lighting conditions, pedestrian postures, and occlusions. A total of 345 pedestrian instances were annotated, with at least one pedestrian in each image and multiple pedestrians in some images. All images were annotated in the PASCAL VOC format, including precise bounding boxes and pixel-level segmentation masks for each pedestrian.