¹Université de Toulon, Aix Marseille Université, CNRS, LIS, UMR 7020, France  ^✉Corresponding Author

Solutions to screen and diagnose positive patients for SARS-CoV-2 promptly and efficiently are critical in the context of the COVID-19 pandemic’s complex evolution. Recent researches have demonstrated the efficiency of deep learning and particularly convolutional neural networks (CNNs) in classifying and detecting lung disease-related lesions from radiographs. This paper presents a solution using ensemble learning techniques on advanced CNNs to classify as well as localize COVID-19-related abnormalities in radiographs. Two classifiers including EfficientNetV2 and NFNet are combined with three detectors, DETR, Yolov7, and EfficientDet. Along with gathering and training the model on a large number of datasets, image augmentation, and cross-validation are also addressed. Since then, this study has shown promising results and has received excellent marks in the Society for Imaging Informatics in Medicine’s competition. The analysis in model selection for the trade-off between speed and accuracy is also given.

• [2023.01.02] Paper is accepted to AICV 2023.
• [2022.09.06] Create baseline.

Evaluation of the COVID-19 lesion detector on the SIIM test set:

Detector	Accuracy (mAP@ IoU 0.5:0.95)			Performance
	Discrete model	Lung localized	Inference speed (FPS)	GPU memory requirement (MB)	Model size (MB)
DETR	0.542	0.587	25	2683	232
Yolov7	0.563	0.591	34	3520	290
EfficientDet	0.499	0.574	19	1903	187
Weighted Box Fusion	0.605	0.612	8	8106	709

Please refer to INSTALL.md for installation instructions.

Trained models are available in the MODEL_ZOO.md.

Please see DATASET_ZOO.md for detailed description of the training/evaluation datasets.

Follow the aforementioned instructions to install environments and download models and datasets.

GETTING_STARTED.md provides a brief intro of the usage of builtin command-line tools.

If you use this work in your research or wish to refer to the results, please use the following BibTeX entry.