This is the unofficial code of Deep Dual-resolution Networks for Real-time and Accurate Semantic Segmentation of Road Scenes. which achieve state-of-the-art trade-off between accuracy and speed on cityscapes and camvid, without using inference acceleration and extra data
我跑了slim版本的是75多，按照论文设置权重的话精度还会降低一些，你呢
Originally posted by @J0kerrr164 in https://github.com/chenjun2hao/DDRNet.pytorch/issues/22#issuecomment-1188930457