def __call__(self, inp, output_prev, _img):
shared = self.Shared(self.Embed(inp))
img_pred = self.Visual(shared)
txt_pred = softmax3d(
self.Embed.unembed(
self.ToTxt(self.LM(last(shared), self.Embed(output_prev)))))
return (img_pred, txt_pred)
def __call__(self, inp, output_prev, img):
img_pred = self.Visual(self.Embed(inp))
txt_pred = softmax3d(
self.Embed.unembed(
self.ToTxt(self.LM(self.FromImg(img),
self.Embed(output_prev)))))
return (img_pred, txt_pred)
def __call__(self, inp, output_prev, _img):
rep = self.Visual.encode(self.Embed(inp))
img_pred = self.Visual.visual_activation(self.Visual.ToImg(rep))
txt_pred = softmax3d(
self.Embed.unembed(
self.ToTxt(self.LM(rep, self.Embed(output_prev)))))
return (img_pred, txt_pred)
def __call__(self, prev):
return softmax3d(self.network.ToTxt(self.network.EncodeLM(self.network.Shared(prev))))
def __call__(self, out_prev):
return softmax3d(self.ToTxt(self.Decode(out_prev)))
def __call__(self, inp, output_prev, _img):
img_pred = self.Visual(self.Embed(inp))
txt_pred = softmax3d(self.ToTxt(self.LM(self.Embed(output_prev))))
return (img_pred, txt_pred)
def __call__(self, inp, output_prev, _img):
img_pred = self.Visual(self.Embed(inp))
txt_pred = softmax3d(self.OH(output_prev)) # fake output prediction
return (img_pred, txt_pred)
def __call__(self, inp, output_prev, _img):
rep = self.Visual.encode(self.Embed(inp))
img_pred = self.Visual.visual_activation(self.Visual.ToImg(rep))
txt_pred = softmax3d(self.Embed.unembed(self.ToTxt(self.LM(rep,
self.Embed(output_prev)))))
return (img_pred, txt_pred)
def __call__(self, inp, output_prev, _img):
shared = self.Shared(self.Embed(inp))
img_pred = self.Visual(shared)
txt_pred = softmax3d(self.Embed.unembed(self.ToTxt(self.LM(last(shared), self.Embed(output_prev)))))
return (img_pred, txt_pred)
def __call__(self, inp, output_prev, _img):
img_pred = self.Visual(self.Embed(inp))
txt_pred = softmax3d(self.OH(output_prev)) # fake output prediction
return (img_pred, txt_pred)
def __call__(self, inp, out_prev, img):
img_out = self.visual_activation(self.ToVis(last(self.Encode(self.Embed(inp)))))
txt_out = softmax3d(self.Embed.unembed(self.PredictT(self.Decode(self.visual_activation(self.FromVis(img)),
self.Embed(out_prev)))))
return (img_out, txt_out)
def __call__(self, inp, out_prev, _img):
rep = last(self.Encode(self.Embed(inp)))
img = self.visual_activation(self.DecodeV(rep))
txt = softmax3d(self.Embed.unembed(self.PredictT(self.DecodeT(rep, self.Embed(out_prev)))))
return (img, txt)
def __call__(self, inp, output_prev, _img):
inp_e = self.Joint(self.Embed(inp))
output_prev_e = self.Embed(output_prev)
img = self.visual_activation(self.Visual(inp_e))
txt = softmax3d(self.Embed.unembed(self.Textual(inp_e, output_prev_e, _img)))
return (img, txt)
def __call__(self, inp, output_prev, img):
inp_e = self.Embed(inp)
output_prev_e = self.Embed(output_prev)
img_pred = self.visual_activation(self.Visual(inp_e))
txt_pred = softmax3d(self.Embed.unembed(self.Textual(inp_e, output_prev_e, img)))
return (img_pred, txt_pred)
