ResDNet: Efficient Dense Multi-Scale Representations With Residual Learning for High-Level Vision Tasks (TNNLS 2022) The official implementation of our paper.