English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

We propose combining memory saving techniques with traditional U-Netarchitectures to increase the complexity of the models on the Brain TumorSegmentation (BraTS) challenge. The BraTS challenge consists of a 3Dsegmentation of a 240x240x155x4 input image into a set of tumor classes.Because of the large volume and need for 3D convolutional layers, this task isvery memory intensive. To address this, prior approaches use smaller croppedimages while constraining the model's depth and width. Our 3D U-Net uses areversible version of the mobile inverted bottleneck block defined inMobileNetV2, MnasNet and the more recent EfficientNet architectures to saveactivation memory during training. Using reversible layers enables the model torecompute input activations given the outputs of that layer, saving memory byeliminating the need to store activations during the forward pass. The invertedresidual bottleneck block uses lightweight depthwise separable convolutions toreduce computation by decomposing convolutions into a pointwise convolution anda depthwise convolution. Further, this block inverts traditional bottleneckblocks by placing an intermediate expansion layer between the input and outputlinear 1x1 convolution, reducing the total number of channels. Given a fixedmemory budget, with these memory saving techniques, we are able to train imagevolumes up to 3x larger, models with 25% more depth, or models with up to 2xthe number of channels than a corresponding non-reversible network.

Memory Efficient 3D U-Net with Reversible Mobile Inverted Bottlenecks for Brain Tumor Segmentation