ConvALSR-Net

Our model code will be uploaded soon.

Data Preprocessing

Please follw the GeoSeg to preprocess the LoveDA, Potsdam and Vaihingen dataset.

Training

"-c" means the path of the config, use different config to train different models.

python train_supervision_dp.py -c ./config/potsdam/convalsrnet.py

python train_supervision_dp.py -c ./config/vaihingen/convalsrnet.py

python train_supervision_dp.py -c ./config/loveda/convalsrnet.py

Testing

Vaihingen

python test_vaihingen.py -c ./config/vaihingen/convalsrnet.py -o ./fig_results/convalsrnet_vaihingen/ --rgb -t "d4"

Potsdam

python test_potsdam.py -c ./config/potsdam/convalsrnet.py -o ./fig_results/convalsrnet_potsdam/ --rgb -t "d4"

LoveDA (Online Testing)

Output RGB images (Offline testing, using the validation set for testing, directly output the mIOU results)

python test_loveda.py -c ./config/loveda/convalsrnet.py -o ./fig_results/convalsrnet_loveda_rgb --rgb --val -t "d4"

Output label images (need to be compressed and uploaded to the online testing website)

python test_loveda.py -c ./config/loveda/convalsrnet.py -o ./fig_results/convalsrnet_loveda_onlinetest -t "d4"

Acknowledgement

Our training scripts come from ConvLSR-Net which is based on GeoSeg. Thanks for the author's open-sourcing code.

• ConvLSR-Net
• GeoSeg(UNetFormer)
• pytorch lightning
• timm
• pytorch-toolbelt
• ttach
• catalyst
• mmsegmentation