FRCD

This is for FRCD: Feature Refine Change Detection Network for Remote Sensing Imagesin remote sensing images.

The code will be polished soon. IF YOU HAVE ANY QUESTIONS, PLEASE DISCUSS THEM IN THE ISSUE PART.

Dataset shodld be in ./dataset/CCD_real

How to train:

python -m torch.distributed.launch --nproc_per_node 2 onlytrain.py

evaluating:

python onlyeval.py

The drawing script will be updated soon.

We will upload the trained model .pth in the future.

The configs can be found in ./configs and

2 is the number of GPUs you want to use. Every GPU have 8 batch size.

The codes are based on SwinTransformer official implementation and MMLab's mmsegmentation and MMCV.

Environment that may used to run the code is as follows:

# Name                    Version                   Build  Channel
cudatoolkit               11.3.1               h2bc3f7f_2    defaults
einops                    0.4.1                    pypi_0    pypi
mmcv-full                 1.4.8                    pypi_0    pypi
mmsegmentation            0.23.0                   pypi_0    pypi
numpy                     1.21.5           py39he7a7128_1    defaults
opencv-python             4.5.5.64                 pypi_0    pypi
pillow                    9.0.1            py39h22f2fdc_0    defaults
python                    3.9.12               h12debd9_1    defaults
pytorch                   1.11.0          py3.9_cuda11.3_cudnn8.2.0_0    pytorch
scikit-image              0.19.2           py39h51133e4_0    defaults
scikit-learn              1.2.2                    pypi_0    pypi
scipy                     1.7.3            py39hc147768_0    defaults
tensorboard               2.8.0                    pypi_0    pypi
timm                      0.6.12                   pypi_0    pypi
tqdm                      4.64.0                   pypi_0    pypi
yacs                      0.1.8                    pypi_0    pypi