What is ACID?
Alberta Construction Dataset Features​
-
10 categories of construction machines
-
10,000 labeled images
-
15,767 construction machine objects
-
Object detection
The Alberta Construction Image Dataset (ACID) is a construction machine detection dataset developed by AIRCon-Lab at the University of Alberta. ACID was developed as a resource to facilitate the use and development of deep learning applications in the construction automation field. The dataset contains images collected from construction sites all over the world and is available for download.
Dataset image examples
Deep Learning Algorithms for Construction
Deep Learning algorithms can automatically identify construction machines in images and videos using parallel computation and graphic card implementation. However, deep learning algorithms require large-scale data sets to avoid the overfitting problem in the training stage. Because construction sites are often difficult to access, it is challenging to create large-scale datasets containing quality images of construction machines. ACID is, therefore, developed as a construction machine dataset to train deep learning algorithms.
​
The ACID team has used the dataset to benchmarked existing deep learning algorithms and develop a set of instructions for selecting proper detectors for construction scenarios. The videos below demonstrate the detection of construction machines bu an algorithm trained by the ACID dataset.
Impacts of ACID
Until now, ACID has been shared with over 400 researchers from universities/companies of 30 different countries.
Cite ACID
The ACID paper (download link) has been published on the Journal of Computing in Civil Engineering. If you find ACID is helpful to your work, please cite the manuscript as follows:
Xiao, B., & Kang, S. C. (2021). Development of an image data set of construction machines for deep learning object detection. Journal of Computing in Civil Engineering, 35(2), 05020005.
​
​