Attention-Guided Pyramid Context Networks

Paper Link
Authors: Tianfang Zhang, Lei Li, Siying Cao, Tian Pu and Zhenming Peng

Datasets

• MDFA dataset is available at MDvsFa cGan.
• SIRST dataset is available at SIRST.
• The SIRST Augment dataset: download from Google Drive or BaiduYun Drive with code ojd4.

Usage

Train

python train.py --net-name agpcnet_1 --batch-size 8 --save-iter-step 20 --dataset mdfa

python train.py --net-name agpcnet_1 --batch-size 8 --save-iter-step 40 --dataset sirstaug

python train.py --net-name agpcnet_1 --batch-size 8 --save-iter-step 100 --dataset merged

Inference

python inference.py --pkl-path {checkpoint path} --image-path {image path}

Evaluation

python evaluation.py --dataset {dataset name} 
                     --sirstaug-dir {base dir of sirstaug}
                     --mdfa-dir {base dir of MDFA}
                     --pkl-path {checkpoint path}

Results

Methods	Data	Precision	Recall	mIoU	Fmeasure	AUC	Download
AGPCNet+ResNet18	MDFA	0.5939	0.7241	0.4843	0.6525	0.8682	model
AGPCNet+ResNet18	SIRST Aug	0.8323	0.8542	0.7288	0.8431	0.9344	model
AGPCNet+ResNet18	Merged	0.7453	0.8384	0.6517	0.7891	0.9194	model

Figure. Illustration of ROC compared with state-of-the-art methods.

Evaluation of model-driven algorithms based on traditional metrics refers [ISTD-python].

Citation

@article{zhang2023attention,
  title={Attention-guided pyramid context networks for detecting infrared small target under complex background},
  author={Zhang, Tianfang and Li, Lei and Cao, Siying and Pu, Tian and Peng, Zhenming},
  journal={IEEE Transactions on Aerospace and Electronic Systems},
  year={2023},
  doi={10.1109/TAES.2023.3238703},
  publisher={IEEE}
}