A pytorch implementation of paper BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation
- Hardware: PC or Server with two NVIDIA 1080Ti GPUs.
- Software: Ubuntu 16.04, CUDA 10.0, Anaconda3, pytorch 1.1.0
Download Cityscapes dataset here or wherever convenient for you. Then run script /datasets/cityscapes/tools/convert_labels.py to generate trainId from labelId.
CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch --nproc_per_node=2 train.py
python evaluate.py
The final mIoU will be around 78.5, depending on random initialization. In order to confirm the experimental results, ckeckpoint (mIoU=78.73) is provided for testing. The examples of final result:
