Abstract

Instance segmentation is more challenging and difficult than object detection and semantic segmentation. It paves the way for the realization of a complete scene understanding, and has been widely used in robotics, automatic driving, medical care, and other aspects. However, there are some problems in instance segmentation methods, such as the low detection efficiency for low-resolution objects and the slow detection speed of images with complex backgrounds. To solve these problems, this paper proposes an instance segmentation method with multi-scale attention, which is called a Hybrid Kernel Mask R-CNN. Firstly, the hybrid convolution kernel is constructed by combining different kernels and groups, which can complement each other to extract rich information. Secondly, a multi-scale attention mechanism is designed by assign weights to different convolution kernels, which can retain more important information. After the introduction of our strategy, the network is more inclined to focus on the low-resolution objects in the image. The proposed method achieves the best accuracy over the anchor-based method. To verify the universality of the model, we test Hybrid Kernel Mask R-CNN on Balloon, xBD and COCO datasets. The test results exceed the state of art methods. And the visualization results show our method can extract low-resolution objects effectively.

Highlights

  • Instance segmentation is a complex issue and one of the most challenging computer vision tasks, which can perform instance segmentation by detecting objects and predicting pixel-level instances on objects.Instance segmentation can be roughly divided into segmentation-based methods and detection-based methods

  • Inspired by the above principles and observations, this paper proposes Hybrid Kernel Mask R-CNN (HKMask)

  • Mask R-CNN [6], one of the detection-based methods, which changes ROI pooling to a quantization-free layer called ROIAlign and generates a binary mask for each class independently. It has achieved the best result of a single model in the 2018 COCO [16] Instance Segmentation Challenge

Read more

Summary

Introduction

Instance segmentation is a complex issue and one of the most challenging computer vision tasks, which can perform instance segmentation by detecting objects and predicting pixel-level instances on objects. Segmentation-based methods predict pixel-level categories and aggregate the same class to achieve the final instance segmentation results. Detection-based methods can trace back to DeepMask [3], which generates instance masks by using sliding windows and predicts the mask by utilizing some detectors, such as R-FCN [4]. Based on Faster R-CNN [7] a fully convolutional network (FCN) is used to achieve semantic segmentation by adding mask branching. Li et al [10] propose TridentNet to generate scale-specific feature maps with a uniform representational power. These methods usually use fixed kernel size, 3×3. For the corresponding kernels, it uses different groups. (2) Based on the Squeeze-and-Excitation Networks, the model provides an improved channel attention module that preserves more important information through the idea of shortcut connection

Instance segmentation based on mask R-CNN
Channel attention module
Hybrid kernel
Attentional mechanism
Experiment
Datasets
Experimental settings
Results and ablation study
Method backbone
Conclusion
Attentional Mechanism
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call