search

MIC-DKFZ / batchgenerators

A framework for data augmentation for 2D and 3D image classification and segmentation
Apache License 2.0
1.1k stars 221 forks source link

10x performance improvement of DownsampleSegForDSTransform2 in nnunet by moving 1 line of code #120

Open AwesomeLemon opened 9 months ago

AwesomeLemon commented 9 months ago

Hi,

Finding unique labels is not necessary when order == 0. In DownsampleSegForDSTransform2 in nnunet, this function is called multiple times, and it all adds up, especially in the limited CPU scenario with high-res images. By introducing this change, I've reduced the time of this transform from ~2 seconds to ~0.2 seconds.

Best,