SpaTUn: Spatio-Temporal Uncertainty
Description.
| Model Type | Likelihood | Examples |
|---|---|---|
| occupancy3d | Bernoulli | 3D continuous occupancy maps |
| scalarfield2d | Gaussian, Gamma | Elevation maps |
| scalarfield3d | Gaussian, Gamma | Position maps, Velocity maps |
| vetorfield3d | Gaussian | Velocity maps, acceleration maps |
| surface3d | Bernoulli | 3D surfaces |
Prerequisites
Install required frameworks:
PyTorch 1.3.0
$ pip3 install torch torchvision
Plotly 4.6.0
$ pip install plotly==4.6.0
Clone the repository
git clone https://github.com/RansML/SpaTUn.git
Example Configurations
The BHM training and plotting module provides a way to quickly save and load experiment configurations via the configs folder. This can be done by setting --config [config_file_name] to load a config, or --save_config [save_to_path] to save a config. For a full list of editable hyper-parameters, refer to the documentation in train.py.
occupancy3d configurations (Vivian and Jason)
Toy3 Dataset: Simple dataset with LIDAR hits along with adjustable uncertainty (sigma). The dataset is located in datasets/toy3/toy3.csv. To run the sample, use:
python3 spatun.py --config toy3_occupancy
scalarfield2d configurations (Lydia, Jason, and Ahmed)
Toy Dataset: Simple dataset with LIDAR hits along with adjustable uncertainty (sigma). The dataset is located in datasets/toy/toy.csv. To run the sample, use:
python3 spatun.py --config toy_regression
Toy2 Dataset: Simple dataset with LIDAR hits along with adjustable uncertainty (sigma). The dataset is located in datasets/toy2/toy2.csv. To run the sample, use:
python3 spatun.py --config toy2_regression
Toy3 Dataset: Simple dataset with LIDAR hits along with adjustable uncertainty (sigma). The dataset is located in datasets/toy3/toy3.csv. To run the sample, use:
python3 spatun.py --config toy3_regression
scalarfield3d/vetorfield3d configurations (Kyle and Ransalu)
Toy 3D velocty dataset in /datasets/toy_velocity/toy_velocity_1.csv:
python spatun.py --config toy_velocity_gauss
Carla radar dataset in /velocity1/radar_carla_test1_frame_250.csv:
python spatun.py --config velocity1_gauss
For Kyle and Ransalu
python spatun.py --config kyle_ransalu/1_toy1_vel --mode tqp
surface3d configurations (Vivian)
bernoulli-surface Dataset: Dataset with samples on the mesh of a shell. The mesh is located in ply_files. To make a csv from sample hit points on the shell, run:
python3 make_csv.py ply_files/shell.ply datasets/bernoulli-surface/shell.csv
Change the parameter 'dataset_path' in the bernoulli-surface config file to the path of the outputted csv file and run:
python3 spatun.py --config bernoulli-surface