Profiles
test
This error will be reported directly:
TypeError: type object got multiple values for keyword argument 'groups'
He said the dictionary is more than a corresponding value, check the configuration file, found that there are two groups
, the value is the same, only to see a profile above the other, comment out the following tried it, it
results:
training
1. The profile change their class reference data set that: configuration changes
Note : Be sure to set optimistic about the learning rate, 16GPU is 0.02, a single card or dual card set around 0.0025, must be small, at first I dual-card set of 0.02, loss soared from single digits to five digits
2. Because I only need to detect bbox and mask, so the model dictionary in the configuration file are related to semantic annotated
and provided train_pineline in with_seg=False
:
Set only with_seg it is of no use, and finally calculates the loss seg, but there is no label, on the error, so to add front of the Notes
3. Modify the assessment content:
4. Leave a TensorboardLoggerHook
comment on the command line will be able to see the loss
but my program will pop up a lot of warning, do the following can not display a warning:
import warnings
warnings.filterwarnings('ignore')
Again after commissioning appear:
Address alredy in use
This shows that memory is not just to stop the release of the program, execute the following command:
ps aux|grep user_name|grep python
And then kill -9 ID
to free memory.
This will successfully trained:
Precautions
resize Here we must pay attention! Because when I was a beginning of a second too large, the results of super memory, our group walked in another program collapse out of seven days. . . .