Known Issues¶
Boundary prediction not starting¶
Relevant for:
- Windows
- nvidia GPU
On Windows using a nvidia GPUs PlantSeg might appear stuck running the prediction. If this error occurs, check the GPU memory in the task manager, it is probably full.
GPUs can use the system memory after running out of memory
on the GPU itself. This prevents crashes and errors, but makes all operations
very slow.
In PlanSeg, this "feature" prevents us from finding optimal patch and batch
sizes for your GPU.
Here is how you can fix the issue:
A) Quick and dirty fix: Set patch and batch size manually¶
In Segmentation
, Boundary Prediction
, click on Show advanced parameters
.
Then check the box Manual Patch Size
and enter your patch size (min 64x64 in x,y dimension).
Set Batch size
to One (low VRAM usage)
.
B) Proper fix: Disable memory off-loading¶
Open your nvidia control panel. Navigate to 3D-Settings
(1), Program specific
(not global)(2).
Click on add
and choose Python, afterwards it should display the path to the python
executable in you PlantSeg installation(3).
Scroll to CUDA - Sysmem Fallback Policy
and change that to Prefer No Sysmem Fallback
(4).