Indoor fire load detection using deep learning-based instance segmentation.
This repo contains the dataset, codes, and documents for the following paper:
Y. -C. Zhou, Z. -Z. Hu, K. -X. Yan and J. -R. Lin, "Deep Learning-Based Instance Segmentation for Indoor Fire Load Recognition," in IEEE Access, vol. 9, pp. 148771-148782, 2021, doi: 10.1109/ACCESS.2021.3124831.
The data/indoor-scene/trainval1k.zip is the dataset containing all images with annotations developed in this research.
-
Install Pytorch 1.6+ and detectron2
-
Clone or download the repo
git clone https://github.com/Zhou-Yucheng/fire-load-detection.git
cd fire-load-detection/src
-
Unzip the dataset trainval1k.zip in data/indoor-scene
-
Run
python3 train.py --help
for more information about usage -
Run train.py with arguments, for example:
python3 train.py -m R50 -b 4 -l 2e-3 -i 6k --step 4k