To the best of our knowledge, we developed and opened the first image dataset for the detection and segmentation of indoor fire loads. With in the dataset, there are 8858 instances in 1015 images labelled according to their materials (such as fabric, wood, plastic, etc.) and combustibility, which are quite useful for indoor fire load estimation, resilience assessment, etc. We also developed a Mask-RCNN-based method for indoor fire load recognition, ref to our paper for details. Source code and dataset are also available on github for future invesigations.
You May Also Enjoy
Together with Prof. van Treeck’s group, we developed a tool for AR-based CFD visualization of indoor thermal environment, which could reduce computational cost and improve visualization efficiency dramatically
With collaboration with Dr. Yichuan Deng from South China University of Technology, we developed a large-scale image dataset for deep learning based site object detection, the dataset contains images taken from different view angles and ambient lighting conditions, and covers various objects such workers, equipments, materials, etc.