Generalize CONV_2D op to support input product scale > output scale
To remove inconsistency with TF Lite. TF Lite removed input/output scale restrictions to support additional use cases. This is a superset of existing CONV_2D functionality. Bug: 127427474 Test: NeuralNetworksTest_static Change-Id: I7d9f7afa0ae177c9169af012b2ba915a1341a698
Loading
Please sign in to comment