Fig. 5From: From compressive sampling to compressive tasking: retrieving semantics in compressed domain with low bandwidthThe architecture of the physics-driven two-stage unrolling model for SCI reconstruction. Each stage in the model consists of a projection step and a prior learning step. By taking advantage of the unrolling design, this model can be trained with small patches and then finetuned to large-scale data efficientlyBack to article page