For convenience, some checkpoints, such as the MAE-pretrained ViT-B model, are provided for manual download. Users must update the following paths accordingly. Relevant checkpoints can be acquired from the website.
-
❗ pretrain.sh, finetune.sh, scratch, eval.sh: Please update the following:
- calvin_dataset_path to the directory where you have stored the CALVIN ABC-D data.
- save_checkpoint_path to the parent directory where your experiment checkpoints are saved. Recommend to create a
checkpoints
folder in the project root directory. - finetune_from_pretrained_ckpt to the location of your pre-trained checkpoint.
- resume_from_checkpoint to the location of your fine-tuned checkpoint.
- vit_checkpoint_path to the location of your ViT checkpoint (downloaded from the website). Recommend to be stored in
checkpoints/vit_mae/mae_pretrain_vit_base.pth
.
-
❗ networkx: Due to compatibility issues between the networkx library in CALVIN and Python 3.10, we provide a compatible version of networkx.zip on the website. Download and unzip it, then replace the existing networkx library in the following path:
# Pre-train Seer on Calvin ABC-D dataset
bash scripts/CALVIN_ABC_D/Seer/pretrain.sh
# Pre-train Seer-Large on Calvin ABC-D dataset
bash scripts/CALVIN_ABC_D/Seer-Large/pretrain.sh
# Fine-tune Seer on Calvin ABC-D dataset
bash scripts/CALVIN_ABC_D/Seer/finetune.sh
# Fine-tune Seer-Large on Calvin ABC-D dataset
bash scripts/CALVIN_ABC_D/Seer-Large/finetune.sh
# Train Seer on Calvin ABC-D dataset from scratch
bash scripts/CALVIN_ABC_D/Seer/scratch.sh
# Train Seer-Large on Calvin ABC-D dataset from scratch
bash scripts/CALVIN_ABC_D/Seer-Large/scratch.sh
# Evaluate Seer on Calvin ABC-D benchmark
bash scripts/CALVIN_ABC_D/Seer/eval.sh
# Evaluate Seer-Large on Calvin ABC-D benchmark
bash scripts/CALVIN_ABC_D/Seer-Large/eval.sh