Running

Notice

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.

• :exclamation: pretrain.sh, finetune.sh, scratch, eval.sh: Please update the following:
  • 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.
  • libero_path to the location of LIBERO dir.

Seer

Convert Data

python utils/convert_libero_per_step.py

Pre-train

# Pre-train Seer on LIBERO-90 dataset
bash scripts/LIBERO_LONG/Seer/pretrain.sh

Fine-tune

# Fine-tune Seer on LIBERO-10 dataset
bash scripts/LIBERO_LONG/Seer/finetune.sh

Train from Scratch

# Train Seer on LIBERO-10 dataset from scratch
bash scripts/LIBERO_LONG/Seer/scratch.sh

Eval

# Evaluate Seer on LIBERO-10 benchmark
bash scripts/LIBERO_LONG/Seer/eval.sh