SOL-SLAM

Submission for TIM, "An Online Adaptation Method for Robust Depth Estimation and Visual Odometry in the Open World".

Install Dependency Packages

conda env create -f requirement.yml -p {ANACONDA_DIR/envs/sol_slam}
conda activate sol_slam

File Description

• apis contains the application program interface scripts.

• libs/sol.py is the main program of the SOL-SLAM system.

• libs/datasets contains the different dataset interface scripts.

• libs/deep_models contains the deep learning models of the program, such as Monodepth2, SC-DepthV3.

• libs/flowlib contains the configuration library of the flow estimation network

• libs/general contains the base classes, such as timer, utils.

• libs/geometry contains the geometric algorithms, such as triangulation, reprojection.

• libs/matching contains the 2D feature selection and matching scripts.

• libs/tracker contains the 2D feature tracking scripts.

• options contains the configuration files.

• tools contains the evaluation scripts.

SOL-SLAM

Online Learning

python apis/run.py -d options/example/default_configuration.yml -c options/examples/DATASET_config/ol.yml --seq SEQ --no_confirm --off_flownet --save_pose

Offline Learning

python apis/run.py -d options/example/default_configuration.yml -c options/examples/DATASET_config/off.yml --seq SEQ --no_confirm --off_flownet --save_pose

Evaluation

python tools/eval_depth_fixed.py --dataset DATASET --pre_depth PREDICTION_PATH --gt_depth GT_PATH