Save GPU RAM Usage in Convolutional Layers to Load Huge Images
Ando, T.
Show abstract
Image recognition models have evolved tremendously. Despite the progress for general images, histopathological images are not easy targets. One of the reasons is that histopathological images can be 100000-200000px in height and width which are often too large for a deep neural network model to handle directly because the RAM of the GPU is limited. Mitigating the obstacle is expected to be a progress in the histopathological image analysis. In this study, we save the RAM consumption in a convolutional layer by allocating only the required data to GPU only when needed and by dividing the calculation into per channel. This RAM Saving Convolutional layer (RSConv) can load larger images than a normal convolutional layer. The code is available at https://github.com/tand826/RAMSavingConv2d.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- A novel interpretable deep transfer learning combining diverse learnable parameters for improved T2D prediction based on single-cell gene regulatory networks 94%
- Estimation of Three-Dimensional Chromatin Morphology for Nuclear Classification and Characterisation 94%
- A Robust Spike Sorting Method based on the Joint Optimization of Linear Discrimination Analysis and Density Peaks 93%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.