This is an implementation of our paper in SSCI 2023. SemanticSLam: Learning based Map Construction and Robust Camera Localization.
pip install -r requirements.txt
pip install torch-scatter -f https://data.pyg.org/whl/torch-1.11.0+cu102.html
Note: To install torch-scatter, specify the cuda version which is depending on your PyTorch installation(e.g.cpu, cu102, cu113, or cu115)
This code is implemented in Python >=3.8.
Download datasets to "dataset" folder
https://drive.google.com/file/d/18J8n4-tKxwI7RJlyu2gGF_S2-9zOHJoU/view?usp=sharing
https://drive.google.com/file/d/11UnOoFUvB2b24M1KiQ40-SMzUIxK-j27/view?usp=sharing
Traing works with default arguments by:
python train.py --savedate SaveName
SaveName is the name of logging file and saving models.
Evaluation can be done as follows:
python evaluate.py --savedate SaveName
SaveName should be same as training to load well-trained modles