Instance segmentation algorithm based on fine-grained feature perception and cross-path aggregation

Most existing instance segmentation methods are based on object detection models, and their segmentation performance is limited by the accuracy of the detection boxes. To overcome the limitation of detection boxes, we propose an instance segmentation algorithm based on fine-grained feature perception and cross-path aggregation. Our approach aims to find strongly distinguished feature regions without relying on detection boxes and to perform precise local segmentation. Therefore, a fine-grained feature perception module is designed in the decoding stage by compressing the number of channels to aggregate scattered feature responses and then using an activation function to obtain a weight matrix. Finally, the weight matrix is used to find feature maps that focus on different mask regions to perform precise local segmentation. In addition, we also design a new cross-path aggregation structure in the encoding stage to reduce the loss of foreground instance information during network propagation. Experiments on the MS-COCO and PASCAL VOC 2012 datasets validate the effectiveness of our proposed algorithm and show that our method is competitive compared with state-of-the-art methods.& COPY; 2023 Elsevier B.V. All rights reserved.