Stair dataset

Published: 9 January 2023| Version 3 | DOI: 10.17632/3jjdm6rn96.3
Contributors:
Chen Wang,
,
,

Description

We provide a stair dataset with fine annotations for stair detection research. The training set contains 2670 images, and the validation set contains 424 images. Each label contains the locations of the two endpoints and the classification (convex/concave) of each stair line. These images are padded and resized to 512x512 to simplify the data loading process. The main sources of our dataset are as follows: First, we relabel the dataset of reference [1] and add it to our dataset. Then, we use a camera to collect stair images from actual scenes at Beihang University, as well as a few stair images from the Great Wall. Finally, we get some stair images from the internet. The annotation form of the dataset is as follows: cls x1 y1 x2 y2/n ... Each stair line is represented by the above five-tuple data, where cls represents the class of the stair line, 0 represents a convex line and 1 represents a concave line. X1 and y1 represent the coordinates of the left endpoint of the stair line, and x2 and y2 represent the coordinates of the right endpoint of the stair line. The label of an image is stored in a text file and associated by the file name. In addition, to objectively evaluate the model on the dataset, we divide the data into daytime data, night data and art data according to their collection conditions, and the detection difficulty also increases in sequence. The list of daytime, night and art datasets are provided in the following files: 'train_daytime.txt', 'train_night.txt', 'train_art.txt', 'val_daytime.txt', 'val_night.txt' and 'val_art.txt'. [1]Patil, U., Gujarathi, A., Kulkarni, A., Jain, A., Malke, L., Tekade, R., Paigwar, K., Chaturvedi, P.: Deep learning based stair detection and statistical image filtering for autonomous stair climbing. In: 2019 Third IEEE International Conference on Robotic Computing (IRC), pp. 159–166 (2019). https://doi.org/10.1109/IRC.2019.00031

Files

Institutions

Beihang University

Categories

Computer Vision, Robotics, Line Detection, Deep Learning

License