Is it possible to load a part of the network from a symbol block e.g. load resnet18 from “…params” and “…symbol.json” and use it in object detection training as backbone network?
net = gluon.nn.SymbolBlock.imports('resnet18-symbol.json', ['data'], param_file='resnet18-0000.params', ctx=mx.gpu())
This would help a lot to make the backbone architectures independent of the python architecture code behind them.
If not, is there another way to load the network from symbol and params to get this to work?