Inf-Net: Automatic COVID-19 Lung Infection Segmentation from CT Images This paper proposes a novel deep learning network, Inf-Net, for automatic segmentation of lung infections in CT images for COVID-19. The network uses a parallel partial decoder to aggregate high-level features and generate a global map. It also employs implicit reverse attention and explicit edge-attention to model boundaries and enhance representations. To address the challenge of limited labeled data, a semi-supervised segmentation framework is introduced, which leverages unlabeled data to improve learning ability. Extensive experiments on the COVID-SemiSeg dataset and real CT volumes show that Inf-Net outperforms most existing segmentation models, achieving state-of-the-art performance. The proposed Inf-Net is designed to first predict coarse areas of infection and then implicitly model boundaries using reverse attention and edge constraint guidance. A semi-supervised version of Inf-Net is also introduced, which uses a random sampling strategy to enlarge the training set with unlabeled data. This framework is effective in improving segmentation accuracy and is suitable for clinical applications where labeled data is scarce. The method is extended to multi-class infection labeling, allowing for the segmentation of different types of lung infections, such as GGO and consolidation. The semi-supervised Inf-Net is shown to perform well in both single-class and multi-class segmentation tasks. The model is also evaluated on real CT volumes, demonstrating its effectiveness in handling non-infected slices and improving segmentation performance. The proposed method is validated through extensive experiments, showing that Inf-Net achieves superior performance in terms of Dice similarity coefficient, Sensitivity, Specificity, and Precision. The semi-supervised approach is shown to be effective in improving performance, with a 5.7% improvement in Dice score. The model is also effective in multi-class segmentation, outperforming existing state-of-the-art models in several metrics. The paper also presents an ablation study, showing that the key components of the model, including the parallel partial decoder, reverse attention, and edge attention modules, are essential for achieving good performance. The model is further evaluated on real CT volumes, demonstrating its effectiveness in handling non-infected slices and improving segmentation performance. The proposed method has the potential to be applied in clinical settings for the assessment of COVID-19, including quantifying infected regions, monitoring disease progression, and mass screening. The model is also effective in detecting objects with low intensity contrast between infections and normal tissues, which is common in natural camouflage objects. Future work includes applying the model to other related tasks, such as polyp segmentation and camouflaged animal detection.Inf-Net: Automatic COVID-19 Lung Infection Segmentation from CT Images This paper proposes a novel deep learning network, Inf-Net, for automatic segmentation of lung infections in CT images for COVID-19. The network uses a parallel partial decoder to aggregate high-level features and generate a global map. It also employs implicit reverse attention and explicit edge-attention to model boundaries and enhance representations. To address the challenge of limited labeled data, a semi-supervised segmentation framework is introduced, which leverages unlabeled data to improve learning ability. Extensive experiments on the COVID-SemiSeg dataset and real CT volumes show that Inf-Net outperforms most existing segmentation models, achieving state-of-the-art performance. The proposed Inf-Net is designed to first predict coarse areas of infection and then implicitly model boundaries using reverse attention and edge constraint guidance. A semi-supervised version of Inf-Net is also introduced, which uses a random sampling strategy to enlarge the training set with unlabeled data. This framework is effective in improving segmentation accuracy and is suitable for clinical applications where labeled data is scarce. The method is extended to multi-class infection labeling, allowing for the segmentation of different types of lung infections, such as GGO and consolidation. The semi-supervised Inf-Net is shown to perform well in both single-class and multi-class segmentation tasks. The model is also evaluated on real CT volumes, demonstrating its effectiveness in handling non-infected slices and improving segmentation performance. The proposed method is validated through extensive experiments, showing that Inf-Net achieves superior performance in terms of Dice similarity coefficient, Sensitivity, Specificity, and Precision. The semi-supervised approach is shown to be effective in improving performance, with a 5.7% improvement in Dice score. The model is also effective in multi-class segmentation, outperforming existing state-of-the-art models in several metrics. The paper also presents an ablation study, showing that the key components of the model, including the parallel partial decoder, reverse attention, and edge attention modules, are essential for achieving good performance. The model is further evaluated on real CT volumes, demonstrating its effectiveness in handling non-infected slices and improving segmentation performance. The proposed method has the potential to be applied in clinical settings for the assessment of COVID-19, including quantifying infected regions, monitoring disease progression, and mass screening. The model is also effective in detecting objects with low intensity contrast between infections and normal tissues, which is common in natural camouflage objects. Future work includes applying the model to other related tasks, such as polyp segmentation and camouflaged animal detection.