Few-shot Semantic Segmentation with Semantic Collaboration Guidance
The single or a small number of multiple prototypes is insufficient to represent the target information in the whole image.A few-shot semantic segmentation with semantic collaboration guidance is proposed.Firstly,feature extractor with shared weights is used to map the image into the deep feature space and filter out the background regions of the target with the help of the ground truth mask of the support image.Vision Transformer is then used to abstract the depth features directly into multiple prototypes representing the target information in fine granularity.On top of this,semantic information of the target class is introduced as an auxiliary learning task.Finally,a non-parametric metric learning module is used to calculate the similarity values between the query features and the prototypes,and the results are used to guide the pixel-level segmentation for unseen novel classes in the query image.The proposed model is evaluated on the PASCAL-5i and COCO-20i datasets,and results show that the proposed model can achieve competitive segmentation performance on both 1-way 1-shot and 1-way 5-shot settings,with better segmentation performance compared to current mainstream methods.
万方数据
维普
