Added a custom FourierUnitJIT class, exportable using the torch.onnx.export converter (should give identical results to the original torch implementation).
Added information about the ONNX model in the README.
You need to add the use_jit boolean config field in config.yaml under generator/resnet_conv_kwargs, like this: config.generator.resnet_conv_kwargs.use_jit = True.
It is turned off by default and should be compatible with the original code.
Changelog:
torch.onnx.export
converter (should give identical results to the original torch implementation).How to activate FourierUnitJIT:
You need to add the
use_jit
boolean config field inconfig.yaml
undergenerator/resnet_conv_kwargs
, like this:config.generator.resnet_conv_kwargs.use_jit = True
.