Open ty2806 opened 4 years ago
I just read the paper again and now my understanding is the additional channel of the heatmap output is a heatmap of face boundery and is used as a part of the adaptive wing loss.
And your model also has a boundery_channels output which I think is the x-, y- bounderies generated from coordinate channels. now my question is what loss is used for this boundery output?
In your code of the model you have
and in evaler.py the last channel isn't used
So I guess it's used somewhere in the loss?