02/27/2024
CLIP-RC have been accepted by CVPR2024
09/27/2022
External Attention have been accepted by TPAMI.
09/19/2022
SegNeXt have been accepted by NeurIPS'2022.
JSeg is a Semantic segmentation toolbox based on MMSegmentation, Jittor and JDet
JSeg environment requirements:
- System: Linux(e.g. Ubuntu/CentOS/Arch), macOS, or Windows Subsystem of Linux (WSL)
- Python version >= 3.7
- CPU compiler (require at least one of the following)
- g++ (>=5.4.0)
- clang (>=8.0)
- GPU compiler (optional)
- nvcc (>=10.0 for g++ or >=10.2 for clang)
- GPU library: cudnn-dev (recommend tar file installation, reference link)
Step 1: Install the requirements
git clone https://github.com/Jittor/JSeg
cd JSeg
python -m pip install -r requirements.txt
If you have any installation problems for Jittor, please refer to Jittor
Step 2: Install JSeg
cd JSeg
# suggest this
python setup.py develop
# or
python setup.py install
If you don't have permission for install,please add --user
.
Please check the dataset_prepare for dataset preparation, The preparation of the dataset comes from MMSegmentation.
JSeg defines the used model, dataset and training/testing method by config-file
, please check the config.md to learn how it works.
We support single-machine single-gpu, single-machine multi-gpu training, multi-machine training is not supported for the time being. Multi-gpu dependence can be referred to here
python tools/run_net.py --config-file=path/to/config --task=train
# For example
# Single GPU
python tools/run_net.py --config-file=project/fcn/fcn_r50-d8_512x1024_cityscapes_80k.py --task=train
# Multiple GPUs
mpirun -n 8 python tools/run_net.py --config-file=project/fcn/fcn_r50-d8_512x1024_cityscapes_80k.py --task=train
We provide an evaluation script to evaluate the dataset. If there is not enough CPU memory, you can save CPU memory by setting --efficient_val
to store the evaluation results in a local file.
python tools/run_net.py --config-file=path/to/config --resume=path/to/ckp --task=val
# For example
python tools/run_net.py --config-file=project/fcn/fcn_r50-d8_512x1024_cityscapes_80k.py --resume=work_dirs/fcn_r50-d8_512x1024_cityscapes_80k/checkpoints/ckpt_80000.pkl --task=val
We provide a test scripts to save the inference results of the data set, which can be saved in the specified location by setting --save-dir
.
python tools/run_net.py --config-file=path/to/config --resume=path/to/ckp --save-dir=path/to/save_dir --task=test
# For example
python tools/run_net.py --config-file=project/fcn/fcn_r50-d8_512x1024_cityscapes_80k.py --resume=work_dirs/fcn_r50-d8_512x1024_cityscapes_80k/checkpoints/ckpt_80000.pkl --save-dir=./ --task=test
We provide a demo that can predict a single picture. For more information, please see here
from jseg.utils.inference import InferenceSegmentor
def main():
config_file = 'project/fcn/fcn_r50-d8_512x1024_cityscapes_80k.py'
ckp_file = 'work_dirs/fcn_r50-d8_512x1024_cityscapes_80k/checkpoints/ckpt_80000.pkl'
save_dir = './'
image = 'cityscapes/leftImg8bit/val/munster/munster_000069_000019_leftImg8bit.png'
inference_segmentor = InferenceSegmentor(config_file, ckp_file, save_dir)
inference_segmentor.infer(image)
if __name__ == "__main__":
main()
- ✔️ ResNet (CVPR'2016)
- ✔️ ResNeXt (CVPR'2017)
- ✔️ MobileNetV2 (CVPR'2018)
- ✔️ ResNeSt (ArXiv'2020)
- ✔️ Vision Transformer (ICLR'2021)
- ✔️ Swin Transformer (ICCV'2021)
- ✔️ ConvNeXt (CVPR'2022)
- ✔️ BEiT (ICLR'2022)
- ✔️ MAE (CVPR'2022)
- ✔️ FCN (CVPR'2015/TPAMI'2017)
- ✔️ PSPNet (CVPR'2017)
- ✔️ DeepLabV3 (ArXiv'2017)
- ✔️ DeepLabV3+ (CVPR'2018)
- ✔️ UPerNet (ECCV'2018)
- ✔️ NonLocal Net (CVPR'2018)
- ✔️ ANN (ICCV'2019)
- ✔️ DANet (CVPR'2019)
- ✔️ CCNet (ICCV'2019)
- ✔️ EMANet (ICCV'2019)
- ✔️ GCNet (ICCVW'2019/TPAMI'2020)
- ✔️ PointRend (CVPR'2020)
- ✔️ SegFormer (NeurIPS'2021)
- ✔️ EANet (TPAMI)
- ✔️ SegNeXt (NeurIPS'2022)
- ✔️ ADE20K
- ✔️ Cityscapes
- ✔️ PASCAL VOC
- ✔️ iSAID
- ✔️ LoveDA
- ✔️ Potsdam
- ✔️ Vaihingen
Website: http://cg.cs.tsinghua.edu.cn/jittor/
Email: jittor@qq.com
File an issue: https://github.com/Jittor/jittor/issues
QQ Group: 761222083
JSeg is currently maintained by the Tsinghua CSCG Group. If you are also interested in JSeg and want to improve it, Please join us!
@article{hu2020jittor,
title={Jittor: a novel deep learning framework with meta-operators and unified graph execution},
author={Hu, Shi-Min and Liang, Dun and Yang, Guo-Ye and Yang, Guo-Wei and Zhou, Wen-Yang},
journal={Science China Information Sciences},
volume={63},
number={222103},
pages={1--21},
year={2020}
}