A mixed-scale dense convolutional neural network for image analysis

Abstract

Abstract-Deep convolutional neural networks have been successfully applied to many image processing problems in recent works. Popular network architectures often add additional operations and connections to the standard architecture to enable training deeper networks. To achieve accurate results in practice, a large number of trainable parameters is often required. Here, we introduce a network architecture based on using dilated convolutions to capture features at different image scales, and densely connecting all feature maps with each other. The resulting architecture is able to achieve accurate results with relatively few parameters and consists of a single set of operations, making it easier to implement, train, and apply in practice, and automatically adapts to different problems. We compare results of the proposed network architecture with popular existing architectures for several segmentation problems, showing that the proposed architecture is able to achieve accurate results with fewer parameters, with a reduced risk of overfitting the training data.

Interactive volumetric segmentation for textile micro‐tomography data using wavelets and nonlocal means

“…Recently, features learned by Deep Convolutional Neural Networks (DCNN) [12,16,33] have outperformed approaches using hand-engineered image attributes. The drawback of DCNN's is in the requirement to curate a statistically large database.…”

Combining Deep Learning and Active Contours Opens The Way to Robust, Automated Analysis of Brain Cytoarchitectonics

“…To handle the different cell sizes we use the recently proposed Mixed-scale Dense (MS-D) architecture by [11] to robustly predict masks of cell centroid regions in 2D and 3D. The basic concept is as follows: Training: Pairs of image stacks (annotated centroids convolved by a spherical kernel and raw data) are fed into MS-D network.…”

“…Cell Localization For the 2D cell localization we used the MS-D architecture, with a width of 8 (multi-scale feature channels), a depth of 8 and a kernel size of 3×3 (see [11] for details). As the loss function we used the 1 − F β score on the binary pixel labels between reference and prediction.…”

Combining Deep Learning and Active Contours Opens The Way to Robust, Automated Analysis of Brain Cytoarchitectonics

“…The proposed method is based on a Fully Convolutional Neural Network for cell localization and a topology-preserving multi-contour segmentation [1] to control smoothness and topology of the segmentation. To handle the different cell sizes we use the recently proposed Mixed-scale Dense (MS-D) architecture by [11] to robustly predict masks of cell centroid regions in 2D and 3D. The basic concept is as follows: Training: Pairs of image stacks (annotated centroids convolved by a spherical kernel and raw data) are fed into MS-D network.…”