I think the code of ConvGRUCell is right.But MAYBE the code of ConvGRU is wrong.
Input of ConvGRU should be an image sequence,so it should have 5 dimensions(batchsize,timestep,channels,height,weight)
But in your ConvGRU code, the input just has 4 dimensions(batchsize,channels,height,weight),as a result, every ConvGRUCell can not process an image sequence, these cells just regard the image sequence as one image with many channels. So, this code doesn't seem to be working as it should.
I hope you can correct me if I misunderstand.
I think the code of ConvGRUCell is right.But MAYBE the code of ConvGRU is wrong. Input of ConvGRU should be an image sequence,so it should have 5 dimensions(batchsize,timestep,channels,height,weight) But in your ConvGRU code, the input just has 4 dimensions(batchsize,channels,height,weight),as a result, every ConvGRUCell can not process an image sequence, these cells just regard the image sequence as one image with many channels. So, this code doesn't seem to be working as it should. I hope you can correct me if I misunderstand.