Machine Categories
The ACID suite provides data, annotations, and algorithm analysis that contains 10 machine categories, which represent the construction machines most commonly used in construction projects. A sample image is provided below for each machine category.










Dataset Download
The ACID team has conducted comprehensive algorithm analysis on object detection, instance segmentation, and image captioning tasks on the ACID dataset suite. The task introduction, annotation examples, and algorithm results are compiled in below pages. We strongly suggest users browse these pages before downloading ACID datasets. ​
Currently, the ACID datasets, annotations, and guidelines are available for the research community. Please click on the Dataset Download button below and fill the google form. In the google form, you can select which type of annotations (e.g., object detection, instance segmentation, and image captioning) do you need. We will send you the download link in one week once you submitted the google form. If you can't open the google form, please send an email to Dr. Bo Xiao (dr.boxiao@gmail.com) directly to make a request. To be noticed, ACID is currently only sharing for research purposes.
​
​
Dataset Statistics


The number of objects and number of images for each type of construction machines
Number of categories per image in the ACID dataset


Number of objects per image in the ACID dataset
Distribution of object size in the ACID dataset

Number of terms in ACID captioning annotation

Top 20 n-grams (n= 2, 3, and 4) in ACID captioning annotation
Contribute to ACID
Our long-term vision is to establish a decentralized AI hub that connects a self-sustaining organizational structure with scientists, developers, users, policy makers, and educators across academia, industry, and governmental agencies to accelerate AI innovation in construction engineering. The hub will include high-quality, accessible, and reproducible AI datasets (e.g., images, 3D point clouds, and sensor data), models (e.g., building information models), algorithms, domain applications, guidelines, policies and strategies, documentations, and learning materials.
To achieve this goal, ACID needs support to develop a more comprehensive dataset for an innovative construction future. We would like to invite more contributors/researchers to this community and expand the ACID dataset suite. If you want to share your datasets via ACID website, we will be able to setup a specific webpage for you to demo your dataset and related findings to increase your exposure to the community. Please contact Dr. Bo Xiao (dr.boxiao@gmail.com) if you want to contribute to ACID, sharing your datasets via our website, or discuss in-depth collaboration.
Copyrights
Website and Annotations
The annotations (i.e. object detection, instance segmentation, and image captioning) in this dataset along with the website are owned by the ACID Group and are licensed under a Creative Commons Attribution 4.0 International License. No commercial usage is allowed.
Images
The ACID Dataset includes some images extracted from the Internet (e.g., YouTube and Google Image) and some images captured by the ACID's research team.
-
The use of the images extracted from the videos on YouTube must abide by the Copyright on YouTube.
-
The use of the images captured by the ACID's research team must be compliant with the Creative Commons Attribution 4.0 International License.