Nice work!
I noticed that you used a deformGRU network. Actually, I have also thought use deformable convolution to further solve the "deformation" problem in weather radar images. A little different with your model, I used deformable convolution in ConvLSTM instead of ConvGRU. However, I did't get better results compared with ConvLSTM. I don’t know if it’s the model or my parameters are not adjusted properly.
So, deos your Deform-ConvGRU has a better performance than ConvGRU?
Nice work!
I noticed that you used a deformGRU network. Actually, I have also thought use deformable convolution to further solve the "deformation" problem in weather radar images. A little different with your model, I used deformable convolution in ConvLSTM instead of ConvGRU. However, I did't get better results compared with ConvLSTM. I don’t know if it’s the model or my parameters are not adjusted properly.
So, deos your Deform-ConvGRU has a better performance than ConvGRU?