Learning to solve TV regularised problems with unrolled algorithms

While this could be done using the synthesis formulation, we demonstrate that this leads to slower performances.