Troubleshooting

Trouble shooting

Before reading further check your user id.

The dockers are known to only work for users with the id 1000. We are working on a solution. In Ubuntu you can check this typing id in a shell.

It runs slow

• check if you ran the initial "sudo modprobe nvidia_uvm"
• Check if GPU is enabled in the dockers "--gpus all"
• Check if GPU is enabled in the software parameters "--gpu 0" meaning use the first GPU
• Use "top"
• Use "nvidia-smi -l"
• Send screenshots of top and nvidia-smi to Rutger

My results of training are not saved:

• Check docker "-v" mappings

Not found: No algorithm worked!

If you get something like:
"Not found: No algorithm worked!"
This is often an error that is actually caused by running out of memory. Set --memory_limit to 0 and try again. Otherwise decrease number of layers and or units and/or batch_size.

Words are not segmented correctly:

• Note/Warning: new HTR-results will include segmentation on a word-level. The coordinates for this are based on interpolation. Do not use these word-level coordinates for anything scientific. It should be good enough for highlighting in the interface, do not use for anything else.