INSTALL.md

Installation & Data Preparetion

Requirements

• Linux
• Python>=3.6.2 and < 3.9
• PyTorch>=1.4
• torchvision (matching PyTorch install)
• CUDA (must be a version supported by the pytorch version)
• OpenCV
• scikit-image
• importlib-resources=5.12.0

Installing GraSS with VISSL

1. Create Enviroment

conda create -n grass_env python=3.8
conda activate grass_env

2. Install PyTorch

conda install pytorch==2.0.0 torchvision==0.15.0 torchaudio==2.0.0 pytorch-cuda=11.7 -c pytorch -c nvidia

Or Visit Pytorch to install

3. Install Apex(optional)

pip install packaging
git clone --recursive https://www.github.com/NVIDIA/apex
cd apex
python3 setup.py install

4. Install opencv, scikit-image and importlib-resources

pip install opencv-python
pip install scikit-image
pip install importlib-resources==5.12.0

4. Install GraSS

Download GraSS source code and switch to the source path for installation:

git clone --recursive https://github.com/GeoX-Lab/GraSS.git
cd GraSS
pip install --progress-bar off -r requirements.txt
pip install classy-vision@https://github.com/Vaczzy/ClassyVision/tarball/master
pip install -e .[dev]

Data Preparetion and Training

1. Prepare Data, e.g.,

(The data is organized with ImageNet style)

SegmentationImage
|_ <potsdam>
|   _ <ssl_train>
|  |  |_ <train>
|  |  |  |_ <img-t1-name>.tif
|  |  |  |_ ...
|  |  |  |_ <img-tN-name>.tif
|  |  |  |_ ...
|   _ <ssl_val>
|  |  |_ <val>
|  |  |  |_ <img-v1-name>.tif
|  |  |  |_ ...
|  |  |  |_ <img-vN-name>.tif
|  |  |  |_ ...
|_ <loveda_urban>
|  |_ <ssl_train>
|  |  |_ <train>
|  |  |  |_ <img-t1-name>.tif
|  |  |  |_ ...
|  |  |  |_ <img-tN-name>.tif
|  |  |  |_ ...
|_ <loveda_rural>
|  |_ <ssl_train>
|  |  |_ <train>
|  |  |  |_ <img-t1-name>.tif
|  |  |  |_ ...
|  |  |  |_ <img-tN-name>.tif
|  |  |  |_ ...

2. Set Data Path

move to dataset_catalog.json and add (e.g. LoveDA Urban):

{
    "loveda_urban":{
        "train":["SegmentationImage/loveda_urban/ssl_train"," "],
        "val":["SegmentationImage/loveda_urban/ssl_val"," "]
    }
}

3. Data Set

There are two ways to set data config:

a) Move to grass_1gpu_resnet_b256.yaml and change DATASET_NAMES to (e.g. Potsdam):

DATASET_NAMES: ["loveda_urban"] # change dataset name here

and training with this config (grass_1gpu_resnet_b256.yaml)

python tools/run_distributed_engines.py config=pretrain/GraSS/grass_1gpu_resnet_b256.yaml

b) Execute directly without modifying the configuration file:

python tools/run_distributed_engines.py config=pretrain/GraSS/grass_1gpu_resnet_b256.yaml \
config.DATA.TRAIN.DATASET_NAMES=["loveda_urban"]

4. Other Config

All settings of GraSS can be set in grass_1gpu_resnet_b256.yaml, including data augumentation, training epoch, optimizer, adn hyperparameters, etc.