Loading a conv's pretrained weights onto its Decoupled version?

In an attempt to reduce the power usage of a given model, I’m replacing some of its conv2d layers with its Depth-wise Separable version which is composed of a Depth Wise conv2d following a PointWise conv2d. It is then still possible to use the pre-trained weights of the original conv2d? If so, how can I go about it?