This repository is an official PyTorch implementation of our paper "ECMNet:Lightweight Semantic Segmentation with Efficient CNN-Mamba Network".
# cityscapes
python train.py --dataset cityscapes --train_type train --max_epochs 1000 --lr 4.5e-2 --batch_size 5
# camvid
python train.py --dataset camvid --train_type train --max_epochs 1000 --lr 1e-3 --batch_size 8
# cityscapes
python test.py --dataset cityscapes --checkpoint ${CHECKPOINT_FILE}
# camvid
python test.py --dataset camvid --checkpoint ${CHECKPOINT_FILE}
only for cityscapes dataset
python predict.py --dataset cityscapes
- Please refer to our article for more details.
| Methods | Dataset | Input Size | mIoU(%) |
|---|---|---|---|
| ECMNet | Cityscapes | 1024x1024 | 70.6 |
| ECMNet | CamVid | 360x480 | 73.6 |