Applications of Capsules Segmentation
Presentation Outline Capsule Segmentation Challenges & Solutions Applications of Segmentation Capsules Lung Tissue, Retinal Vessels, Handling Transformations. Code Demo University of Central Florida | Center for Research in Computer Vision (CRCV)
Capsule Segmentation Challenges Typically larger image sizes & dense output Required more GPU memory Balance global & local information Requires even more GPU memory University of Central Florida | Center for Research in Computer Vision (CRCV)
Overcoming the Memory Burden Locally-constrained dynamic routing Transformation matrix sharing “Deconvolutional” capsules Encoder-decoder networks University of Central Florida | Center for Research in Computer Vision (CRCV)
Locally-Constrained Dynamic Routing University of Central Florida | Center for Research in Computer Vision (CRCV)
Transformation Matrix Sharing Convolutional capsule layers use different transformation matrices for each member of the grid as well as for each type of capsule. Sharing transformation matrices across members of the grid can further reduce parameters. Still have different matrices for each capsule type. University of Central Florida | Center for Research in Computer Vision (CRCV)
“Deconvolutional” Capsules Prediction vectors formed using transposed convolutions. Routing is computed the same. University of Central Florida | Center for Research in Computer Vision (CRCV)
Example Segmentation Architectures University of Central Florida | Center for Research in Computer Vision (CRCV)
University of Central Florida | Center for Research in Computer Vision (CRCV)
Example Applications Retinal Blood Vessels Pathological Lungs Ground-glass opacity Nodule Emphysema University of Central Florida | Center for Research in Computer Vision (CRCV)
Lung Segmentation University of Central Florida | Center for Research in Computer Vision (CRCV)
Retinal Vessel Segmentation University of Central Florida | Center for Research in Computer Vision (CRCV)
Handling Transformations Credit to Cheng-Lin Li: https://cheng-lin-li.github.io/SegCaps/ University of Central Florida | Center for Research in Computer Vision (CRCV)
Handling Transformations R. Varghese, S. Sharma and M. Premalatha, "Transforming Auto-Encoder and Decoder Network for Pediatric Bone Image Segmentation using a State-of-the-art Semantic Segmentation network on Bone Radiographs," 2018 International Conference on Intelligent Informatics and Biomedical Sciences (ICIIBMS), Bangkok, 2018, pp. 251-256. Handling Transformations Input Image U-Net Output SegCaps Output University of Central Florida | Center for Research in Computer Vision (CRCV)
Code Demo Google Colab Notebook https://drive.google.com /drive/folders/1MhebBrD sh3N5HSntj2Zl5edx56_IkXk N?usp=sharing Can download and use Jupyter Notebook University of Central Florida | Center for Research in Computer Vision (CRCV)
Code is Publically Available Thank You! Questions and Discussions Comment on project page Email lalonde@knights.ucf.edu Project Page https://goo.gl/ySiQHF Code is Publically Available University of Central Florida | Center for Research in Computer Vision (CRCV)