Currently implemented INNs for pytorch. User can concatenate multiple coupling layers to compose an INN.
User can set backward = True and compute input feature map with the existing network from the output feature map.
Current Drawback : input and output shape for each coupling layer is identical (bijectivity is important for inversion).
Currently implemented INNs for pytorch. User can concatenate multiple coupling layers to compose an INN. User can set backward = True and compute input feature map with the existing network from the output feature map. Current Drawback : input and output shape for each coupling layer is identical (bijectivity is important for inversion).