GPU Options
In the GPU Setup panel define the parameters for the used graphic card(s).
Batch Size
The Batch Size is the number of samples, which the training uses simultaneously. For example, with the default batch size 4, during training the error is computed, and the network updated for four windows at the same time, before moving on to the next four windows. A value of 1 slightly decreases training efficiency and can result in noisier training. For example, if a window contains an uncommon geometric configuration, the training could take a step in the wrong direction when computing the updated network weights for the next step. In literature there is usually a warning about too large values for batch size due to the potential of overfitting, but in 3D image analysis here the batch size is limited by the available GPU memory. If the value is chosen too high, a warning dialog appears, when starting the training. It is recommended to set the value as high as possible with the available GPU memory, so that the training works without giving an error. This is often a value between 1 and 16 depending on the GPU hardware.
If multiple GPUs are selected, the batches are distributed to equal numbers on the different GPUs.
|
Select GPUS to use
If a suitable graphics card is detected during installation of GeoDict, the GPU mode is used for running the training, otherwise it is running in CPU mode.
If multiple GPUs are available, select on which it should run. The training can also run on multiple GPUs. If selecting more GPUs than licensed, an error message appears when clicking Run.
|