Official implementation of CVPR21 "Digital Gimbal: End-to-end Deep Image Stabilization with Learnable Exposure Times"
pip3 install -r requirements.txt
python3 run.py -nw <num_cpu_workers> -c <config_file> -r
Drop -r if you resume training from a previous checkpoint.
The configuration file determines the settings of the model, as well as the locations of the training/test datasets and checkpoint directory.
We supply config.ini, which is the main configuration file used for producing the synthetic experiments in the paper.
python3 run.py -nw <num_cpu_workers> -c <config_file> -e
The configuration file used here must match the one used during training, with a valid checkpoint located in the ceckpoint directory.
To reproduce our synthetic experiments, download our pre-trained model using this link, place the checkpoint file inside the checkpoint directory and use config.ini as your configuration file.