What does "--logtostderr" mean in the command line while using tensorflow's object detection api?

Question

when training object detection models using tensorflow, we always input

python train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/ssd_mobilenet_v1_pets.config

But I wonder what the functionality of "--logtostderr" is ? what if omit it?

Answer 1

as the name implies, it sends the logs to STDERR standard file, which would allow you to append at the end of the command: 2> somefilecontainingthelogs.txt

You can read more about STDIN, STDOUT and STDERR here: http://www.learnlinux.org.za/courses/build/shell-scripting/ch01s04.html

If you were to not include the --logtostderr parameter, the logs would typically be sent over to STDOUT; practically if you were to run the command as you have in your question, the result would be the same. But if you were using 2> for redirecting the logs to a file, omitting the --logtostderr would no longer log anything and the logs would appear on the screen since STDOUT is not redirected to a file.