scripts/train_vae/README.md

Train VAE from scratch on SEVIR-LR (downsampled low-resolution version) frames

Commands for training

Run the following command to train VAE on SEVIR-LR frames. Change the configurations in corresponding cfg.yaml

cd ROOT_DIR/PreDiff
MASTER_ADDR=localhost MASTER_PORT=10001 python ./scripts/vae/sevirlr/train_vae_sevirlr.py --gpus 2 --cfg ./scripts/vae/sevirlr/cfg.yaml --save tmp_vae_sevirlr

Or run the following command to directly load pretrained checkpoint for test.

MASTER_ADDR=localhost MASTER_PORT=10001 python ./scripts/vae/sevirlr/train_vae_sevirlr.py --gpus 2 --pretrained --save tmp_vae_sevirlr

Run the tensorboard command to visualize the experiment records

cd ROOT_DIR/PreDiff
tensorboard --logdir ./experiments/tmp_vae_sevirlr/lightning_logs