Skip to content

Latest commit

 

History

History
85 lines (55 loc) · 1.81 KB

TRAIN.md

File metadata and controls

85 lines (55 loc) · 1.81 KB

Train

Image and Raymap VAEs

Train Image VAE

Train the image VAE with one GPU.

./fillerbuster/scripts/dist_run.sh train-image-vae 1 fillerbuster/scripts/train.py

Train the image VAE with SLURM.

python fillerbuster/scripts/slurm.py --method train-image-vae

Train Pose VAE

Train the pose VAE with one GPU.

./fillerbuster/scripts/dist_run.sh train-pose-vae 1 fillerbuster/scripts/train.py

Train the pose VAE with SLURM.

python fillerbuster/scripts/slurm.py --method train-pose-vae
Fillerbuster

Train Fillerbuster

Train the diffusion model with one GPU.

./fillerbuster/scripts/dist_run.sh debug 1 fillerbuster/scripts/train.py

Debug with some modified configs, e.g., a specified seed.

./fillerbuster/scripts/dist_run.sh train 1 fillerbuster/scripts/train.py "--global-seed=52"

Train the diffusion model with SLURM. Run --help to see more options. This command will make a config and print out the RSC command you need to run.

python fillerbuster/scripts/slurm.py --method train

Train with some modified configs, e.g., a specified seed.

python fillerbuster/scripts/slurm.py --method train --extra="--global-seed=52"

Resume from the latest checkpoints specified in the config.

python fillerbuster/scripts/slurm.py --method train-from-checkpoints

Train with higher resolution and change the mask distribution (starting from the train-from-checkpoints config).

python fillerbuster/scripts/slurm.py --method finetune

Test

Run inference with the model from checkpoints.

./fillerbuster/scripts/dist_run.sh train-from-submission-checkpoints 1 fillerbuster/scripts/train.py "--validation-only=True"