Fig. 1

The main framework of our proposed AcneDGNet and its evaluation. (a) is the framework of our proposed AcneDGNet. The input facial image is firstly passed through the feature extraction module, which includes Swin Transformer architecture and feature pyramid architecture. Then the multi-scaled feature maps output by the feature extraction module are input into the lesion detection module and the severity grade module, respectively. In the lesion detection module, the feature maps of acne candidate regions are obtained by region proposal network architecture to predict the ___location and category of each acne lesion in the image. In the severity grading module, the multi-scaled feature maps are resized and combine with the regional lesion-aware feature maps from lesion detection module to predict the severity grade of the acne image. (b) is the evaluation framework for the performance of AcneDGNet. We designed two application scenarios, including online scenario with smartphone and offline scenario with digital camera and VISIA system. Different datasets were selected for the corresponding evaluation purposes with different evaluation methods.