Detectron2 parameters train/eval/checkpoint etc

The documentation about default confg covers all the parameters’ meanings and can be used as reference for that! detectron2.config — detectron2 0.5 documentation

And me dreaming up cfg.MODEL.CHECKPOINT_PERIOD is exactly what they wanted to avoid by disallowing the creation of new keys.

Highlights:

# Number of images per batch across all machines. This is also the number
# of training images per step (i.e. per iteration). 
_C.SOLVER.IMS_PER_BATCH = 16

Phone disk space & Telegram cache

For the second time, discovered that Telegram Cache takes 40gb of disk space.

In the phone’s own menus related to disk space, this was shown as “Pictures” taking 40gb, not the Telegram app and its cache. But this info is exposed through Telegram’s own menus.