Currently if we have a layer X that has query_config.max_matches > 1 then they are stored into these layers named X, X.1, X.2, etc.
In the model configuration file you need to refer to them as X, X.1, X.2, etc.
This is all okay but during training, when the dataset sees "X.1" it looks for that layer in the dataset config file, meaning we need "X.1" and "X.2" and so on to then be defined in the dataset config.
So at the very least we should make it so that it auto-detects that X.1 should just be X. Which could be as simple as reserving "." in the layer name and splitting it. Or could see if there are other problems that can be solved at the same time (I don't know if anything more is necessary actually).
Currently if we have a layer X that has
query_config.max_matches > 1then they are stored into these layers named X, X.1, X.2, etc.In the model configuration file you need to refer to them as X, X.1, X.2, etc.
This is all okay but during training, when the dataset sees "X.1" it looks for that layer in the dataset config file, meaning we need "X.1" and "X.2" and so on to then be defined in the dataset config.
So at the very least we should make it so that it auto-detects that X.1 should just be X. Which could be as simple as reserving "." in the layer name and splitting it. Or could see if there are other problems that can be solved at the same time (I don't know if anything more is necessary actually).