Eventer watchdog / simultaneous training / bug fixing
We have three not very visible but important improvements:

– Under certain conditions runtime errors in the eventer would make the eventer stop working, resulting in no more alarms, emails etc. Now the watchdog will detect such a situation, shut down the eventer and relaunch it. The errors are still noted in the logfile.

– There are now more than one training process on the training servers available, provided that the GPU has enough RAM. In our training server in Eygelshoven we have enough space for 5 training jobs at hte same time. That should be enough to avoid queues in front of the trainer. We recommend to upgrade your local server (raspi etc.) also, because for taking profit from this, you need to have version 1.6.3 on your raspi. Otherwise you will only have trainer process number 1 available.

– A bug in training was fixed: Under some conditions the trainer did not take in all new images to the training, sometimes resulting in poor adaptation to the last added images. This is fixed now.