The official implementation of "Deep Dual-resolution Networks for Real-time and Accurate Semantic Segmentation of Road Scenes"
Achieve state-of-the-art trade-off between accuracy and speed on cityscapes and camvid, without using inference acceleration (like tensorRT) and extra data (like Mapillary)!
The overall architecture of our methods.
The details of "Deep Aggregation Pyramid Pooling Module(DAPPM)".
Currently, this repo contains the model codes and pretrained models for classification and semantic segmentation.
You can refer to HRNet-Semantic-Segmentation-pytorch-v1.1 or directly use the third-party implementation for training and testing our models locally. Thanks for their works!
We will release the whole train and test codes later.
There are some basic training tricks you should employ to reproduce our results including class balance sample, ohem, crop size of 1024x1024. More details can be found in the paper. And there is usually some variation with Cityscapes val results of the same model, maybe about 1% mIoU.
Keep "align_corners=False" in all places if you want to use our pretrained models for evaluation directly.
DDRNet_23_slim on ImageNet(top-1 error:29.8): googledrive
DDRNet_23_slim using timm library, maybe helpful to train on own datasets(top-1 error:26.3, trained with a batch size of 256, warmup, cosine learning rate, 300 epoches and label smoothing): googledrive
DDRNet_23 on ImageNet(top-1 error:24.1): googledrive
DDRNet_39 on ImageNet(top-1 error:22.7): googledrive
DDRNet_23_slim on Cityscapes(val mIoU:77.8): googledrive
DDRNet_23 on Cityscapes(val mIoU:79.5): googledrive
DDRNet_23_slim: 77.4
DDRNet_23: 79.4
DDRNet_39: 80.4 81.9(multi-scale and flip)
DDRNet_39 1.5x: 82.4(multi-scale and flip)
Evaluate the inference speed on Cityscapes dataset.
python3 DDRNet_23_slim_eval_speed.py
DDRNet-23-slim can achieve above 130fps by using the tool.
If you find this repo is useful for your research, Please consider citing our paper:
@article{hong2021deep,
title={Deep Dual-resolution Networks for Real-time and Accurate Semantic Segmentation of Road Scenes},
author={Hong, Yuanduo and Pan, Huihui and Sun, Weichao and Jia, Yisong},
journal={arXiv preprint arXiv:2101.06085},
year={2021}
}