GPU memory error

[nicastel]

Random error
observed once while running on colab T4 instance

[nicastel]

It seems that the Starnet v1 tensorflow 2 python code does not free up the gpu memory => I have forced memory release via numba and I can see the effect on the colab GPU montoring but the issue is still present.

[nicastel]

The issue is present with large images. It seems more related to way I run the SCUNet model and I manage GPU memory. I use the same library as the chaiNNer tool but chaiNNer implements tiling to avoid this issue.

[nicastel]

chaiNNer did encounter a similar issue with SCUNet and solved it via tiling : chaiNNer-org/chaiNNer#2269. I guess tiling is the only viable option