preprocessing.json

{
  "camera_keys": [
    "cam_high",
    "cam_left_wrist",
    "cam_right_wrist"
  ],
  "input_rgb_dtype": "uint8",
  "input_rgb_order": "RGB",
  "training_raw_shape": [
    3,
    240,
    320
  ],
  "vggt_image_size": 256,
  "vggt_spatial_preprocess": "fastwam.datasets.lerobot.processors.fastwam_processor.prepare_vggt_images",
  "vggt_value_range": [
    0,
    1
  ],
  "action_dim": 14,
  "state_dim": 14,
  "action_horizon": 32,
  "replan_steps": 24,
  "normalization": "z-score; use included dataset_stats.json",
  "stats_sha256": "9352f5a0c6fa06183f0b72c86c645b1f952023791e07e426c60228db229f1e17",
  "reconstruction_target": "future",
  "future_frame_offset": 32,
  "random_augmentation_in_inference": false,
  "context_len": 128,
  "language_mask": "original tokenizer padding mask; distinct from action branch all-ones mask"
}