Step 0: Install Miniconda.

Step 1: Create a virtual environment named stt and activate it.

conda create -n stt python=3.10 -y
conda activate stt

Step 2: Install PyTorch2.1.x.

Linux/Windows:

pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url https://download.pytorch.org/whl/cu121

Or

conda install pytorch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 pytorch-cuda=12.1 -c pytorch -c nvidia

Step 4: Install other dependencies.

pip install -U pandas opencv-python tqdm scikit-image einops matplotlib

• Image and label download address: WHU.

You can also choose other sources to download the data, but you need to organize the dataset in the following format:

${DATASET_ROOT} # Dataset root directory, for example: /home/username/data/WHU
├── train
│   ├── img
│   └── label
├── val
│   ├── img
│   └── label
└── test
    ├── img
    └── label

Note: In the project folder, we provide a folder named Data, which contains an example of the organization of the dataset.

• Image and label download address: INRIA.
• You need to use the script 数据切分 to cut the data into a fixed size.

You can also choose other sources to download the data, but you need to organize the dataset in the above format.

If you want to use other datasets, you can refer to the above method to prepare the dataset.

• Use the script GetTrainValTestCSV.py to generate training, validation, and test csv files.
• Use the script GetImgMeanStd.py to get the mean and standard deviation of the training set images.

Parameter Parsing

• line3：os.environ['CUDA_VISIBLE_DEVICES'] = '0, 1, 2, 3' Set the GPU to be used, generally no need to modify.
• line17：backbone: The type of the backbone network, generally no need to modify.
• line22：top_k_s: The number of spatially sparse tokens selected, generally no need to modify.
• line23：top_k_c: The number of channel sparse tokens selected, generally no need to modify.
• line28：BATCH_SIZE: The batch size of a single card, needs to be modified according to the size of the memory.
• line31：DATASET: The path of the training data list, needs to be modified according to the path of the dataset.
• line32：model_path: The save path of the model, generally no need to modify.
• line33：log_path: The save path of the log, generally no need to modify.
• line35：IS_VAL: Whether to verify, generally no need to modify.
• line37：VAL_DATASET: The path of the validation data list, needs to be modified according to the path of the dataset.
• line39：IS_TEST: Whether to test, generally no need to modify.
• line40：TEST_DATASET: The path of the test data list, needs to be modified according to the path of the dataset.
• line45：PRIOR_MEAN: The mean of the training set images, needs to be modified according to the mean of the dataset.
• line46：PRIOR_STD: The standard deviation of the training set images, needs to be modified according to the standard deviation of the dataset.
• line53：load_checkpoint_path: Whether to load the checkpoint, generally empty.
• line55：resume_checkpoint_path: Whether to resume training, generally empty.