Remote Sensing Semantic Segmentation

Semantic segmentation on remote sensing data sets with the mmsegmentation of OpenMMLab 2.0.

Explore benchmark for comparison.

Personal maintenance and use.

Support Dataset

• Potsdam

Support Model

• FCN
• DeepLab
• UNet
• UperNet

Prepare Env

# create env
conda create -n mmseg_rs python=3.8
conda activate mmseg_rs

# install pytorch
conda install -c pytorch pytorch torchvision
# or
pip3 install torch torchvision

# install mmseg of OpenMMLab 2.0
pip install openmim
mim install mmengine 'mmcv>=2.0.0rc0' 'mmsegmentation>=1.0.0rc0'

Launch Experiments

Using mim to deal with experiments.

# training
mim train mmseg {config}  # train on single gpu
mim train mmseg {config} --cfg-options train_dataloader.batch_size=1  # modify cfg options
mim train mmseg {config} --gpus {num_gpus} --launcher pytorch  # train on distributed mode

# testing
mim test mmseg {config} --checkpoint {checkpont}  # test on single gpu
mim test mmseg {config} --checkpoint {checkpont} --gpus {num_gpus} --launcher pytorch  # test on distributed mode

Results

Segmentor	OA	mFscore	mIoU