- Install the original pytorchSLAYER package
- Install EXODUS from the sinabs-exodus folder
- Install additional requirements:
pip install -r requirements.txt
We use PyTorch lightning to set up flexible experiments. Enter one of the example lines below on the command line to start an experiment and a tensorboard instance will automatically be started, which logs to './lightning_logs'. All the training scripts supportcommand line arguments for controlling various aspects of the training procedure. Run the script with --help as argument to see a list of all available options.
Additional arguments to those listed in help, such as max_epochs can be provided and will be passed directly to pytorch lightning Trainer class. See pytorch lightning documentation for further details.
The --gpus=1 option ensures that the models are executed on GPU.
python dvs_gesture/train_dvs_gesture.py --gpus=1 --max_epochs=100
Alternatively parameter sweeps can be run with the following command:
python dvs_gesture/sweep_dvs_gesture.py
python hsd/train_hsd.py --gpus=1
python ssc/train_ssc.py --gpus=1