Behavior recognition is one of the most popular fields in computer vision, and graph convolution network for skeleton based data has been widely used in the field of action recognition, and has achieved remarkable results. Most of the previous studies focused on modeling the relationship between adjacent joint points in natural state, while ignoring the graph topology between non adjacent joint points. In this work, we propose a new graph convolution with EM dynamic routing to learn different structural features dynamically in action recognition based on human skeleton data, and cluster multiple joint points into corresponding graph topologies effectively. The proposed EMD-GC takes the initially artificially defined shared topology as the general prior knowledge of the model, then models the topology according to the specific correlation of each channel, and finally clusters the features into the corresponding topology through the GMM model. Experiments show that EMD-GCN is superior to the most advanced methods on NTU RGB+D dataset.
Colorectal cancer is one of the most common malignancies that can develop from high-risk colon polyps. Currently, the application of deep learning to colon polyp segmentation has obtained high accuracy, which can have good detection rate for small polyps, but due to the limitation to the morphological diversity of polyps and unclear boundary between polyps and surrounding mucosa, there will be insufficient accuracy and unclear segmentation of marginal regions. To address the above problems, we propose Shifted Windows Parallel Network (SWPNet) for real-time accurate polyp segmentation. We design a four-layer Swin-Transformer module with window segmentation as the backbone network, which have strong feature extraction capability, and on top of the improved version of U-Net structured network with additional encoders and decoders, we add Channel Feature Pyramid (CFP) blocks on the deepest three layers of feature extraction, set different rates of dilated convolution respectively, and fuse more scales of semantic information to refine edge segmentation. We conducted experiments on each of the five classical polyp segmentation datasets and achieved high accuracy, especially on the challenging ETIS dataset with 74.1% mean Dice and 65.6% mean IoU respectively.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.