AECIF-Net

This is the implementation of the paper "Attention-Enhanced Co-Interactive Fusion Network (AECIF-Net) for automated structural condition assessment in visual inspection".

Network Architecture

Getting Started

Installation

• Install the required dependencies in requirements.txt.
• Clone this repo:

git clone https://github.com/itschenyu/AECIF-Net.git
cd AECIF-Net

Dataset

• Please download the SBCIV dataset from here and then place it in ./VOCdevkit/VOC2007/.

Pre-trained Weight

• Please download pre-trained weights on Cityscapes from here and place it in ./model_data/.

Model Download

Model	mIoU_Element	mIoU_Defect	Weight
AECIF-Net	92.11	87.16	Link

Training

python train.py

Testing

Evaluating the model on the test set:

python get_miou.py

Inference

Place the inference images in ./img/, and then run:

python predict.py

Citation

If AECIF-Net and the SBCIV dataset are helpful to you, please cite them as:

@article{ZHANG2024105292,
      title = {Attention-Enhanced Co-Interactive Fusion Network (AECIF-Net) for automated structural condition assessment in visual inspection},
      journal = {Automation in Construction},
      volume = {159},
      pages = {105292},
      year = {2024},
      issn = {0926-5805},
      doi = {https://doi.org/10.1016/j.autcon.2024.105292},
      url = {https://www.sciencedirect.com/science/article/pii/S0926580524000281},
      author = {Chenyu Zhang and Zhaozheng Yin and Ruwen Qin}
}

Note

Part of the codes are referred from MTL-Bridge-Inspection project.

The images and corrosion annotations in the dataset are credited to Corrosion Condition State Semantic Segmentation Dataset and COCO-Bridge Dataset.