Work

A Mixed-Mode Acceleration via Sparsity-Adjustable Pruning for Balancing Computation Density in Lightweight CNNs