Open jdaaph opened 5 years ago
In addition to that, I find that it is not easy for us to import the computation graph in our current scheme unless either
Take the squeeseseg
module that you have implemented as an example
The example config file goes
graph:
graph_name: SqueezeSegNet
ang_filter_coef: 0.02
ang_theta_a: [.9, .9, .6, .6]
bi_filter_coef: 0.1
bilateral_theta_a: [.9, .9, .6, .6]
bilateral_theta_r: [.015, .015, .01, .01]
...
And the definition of the referenced SqueezeSegNet
is in the graph.py
file within the module,
making it hard for the outer layer to manage -- config.yml
is read by the top-level module and there is no way for it to know where to find SqueezeSegNet
unless it imports everything. Technically, config.yml
can be anywhere.
I think it will be better if we make a shared folder that contains only the general computation graphs,
or we can specify the top-level module name and make it a convention to put the computation graph in the graph.py
file under it.
As of now we need to import the class definition manually in cli so that all class definition is called before cli logic is run, how do we do this automatically