Timing + memory usage

Training time and memory usage shown across various GPUs, for 2 labeled images and default 100 epochs:

The inference runtime and memory usage increases with the data size. The runtimes shown below are for a single image run once on an RTX 4070S or an A100 with a batch_size of 32. The runtime will be faster if you run many images of the same size input as an array into Cellpose with a large batch_size. The runtimes will also be slightly faster if you have fewer cells/cell pixels.

Table for 2D (RTX 4070S):

Table for 2D (A100):

Table for 3D (A100):

If you are running out of GPU memory for your images, you can reduce the batch_size parameter in the model.eval function or in the CLI (default is 8).

If you have even larger images than above, you will want to tile them before running Cellpose.