feat(mm as default): adding a global DEFAULT_DEPTH_UNIT variable setting mm as default depth unit

src/lerobot/annotations/steerable_pipeline/frames.py

@@ -36,7 +36,7 @@ from typing import Any, Protocol
 import PIL.Image
 import torch
 
-from lerobot.configs.video import VideoEncoderConfig
+from lerobot.configs import VideoEncoderConfig
 from lerobot.datasets.video_utils import decode_video_frames, reencode_video
 
 from .reader import EpisodeRecord, snap_to_frame

src/lerobot/configs/__init__.py

@@ -33,6 +33,7 @@ from .types import (
     RTCAttentionSchedule,
 )
 from .video import (
+    DEFAULT_DEPTH_UNIT,
     VALID_VIDEO_CODECS,
     VIDEO_ENCODER_INFO_KEYS,
     DepthEncoderConfig,
@@ -67,6 +68,7 @@ __all__ = [
     # Factories
     "encoder_config_from_video_info",
     # Constants
+    "DEFAULT_DEPTH_UNIT",
     "VALID_VIDEO_CODECS",
     "VIDEO_ENCODER_INFO_KEYS",
 ]

src/lerobot/configs/default.py

@@ -19,6 +19,8 @@ from dataclasses import dataclass, field
 from lerobot.transforms import ImageTransformsConfig
 from lerobot.utils.import_utils import get_safe_default_video_backend
 
+from .video import DEFAULT_DEPTH_UNIT
+
 
 @dataclass
 class DatasetConfig:
@@ -38,9 +40,9 @@ class DatasetConfig:
     # When True, RGB video frames are returned as uint8 tensors (0-255) instead of float32 (0.0-1.0).
     # This reduces memory and speeds up DataLoader IPC. The training pipeline handles the conversion.
     return_uint8: bool = False
-    # Physical unit depth maps are dequantized to at load time: "mm" (millimetres) or "m" (metres).
+    # Physical unit depth maps are dequantized to at load time: "mm" (millimeters) or "m" (metres).
     # Has no effect on datasets without depth cameras.
-    depth_output_unit: str = "mm"
+    depth_output_unit: str = DEFAULT_DEPTH_UNIT
     streaming: bool = False
 
     def __post_init__(self) -> None:

src/lerobot/configs/video.py

@@ -62,6 +62,7 @@ DEFAULT_DEPTH_MAX: float = 10.0
 DEFAULT_DEPTH_SHIFT: float = 3.5
 DEFAULT_DEPTH_USE_LOG: bool = True
 DEFAULT_DEPTH_PIX_FMT: str = "gray12le"
+DEFAULT_DEPTH_UNIT = "mm"
 
 # Depth-specific tuning fields persisted under ``features[*]["info"]`` as ``video.<name>``.
 DEPTH_ENCODER_INFO_FIELD_NAMES: frozenset[str] = frozenset({"depth_min", "depth_max", "shift", "use_log"})

src/lerobot/datasets/dataset_reader.py

@@ -22,7 +22,7 @@ from pathlib import Path
 import datasets
 import torch
 
-from lerobot.configs.video import DepthEncoderConfig
+from lerobot.configs import DEFAULT_DEPTH_UNIT, DepthEncoderConfig
 
 from .dataset_metadata import LeRobotDatasetMetadata
 from .depth_utils import dequantize_depth
@@ -54,7 +54,7 @@ class DatasetReader:
         delta_timestamps: dict[str, list[float]] | None,
         image_transforms: Callable | None,
         return_uint8: bool = False,
-        depth_output_unit: str = "mm",
+        depth_output_unit: str = DEFAULT_DEPTH_UNIT,
     ):
         """Initialize the reader with metadata, filtering, and transform config.
 

src/lerobot/datasets/lerobot_dataset.py

@@ -24,7 +24,7 @@ import torch.utils
 from huggingface_hub import HfApi, snapshot_download
 from huggingface_hub.errors import RevisionNotFoundError
 
-from lerobot.configs import DepthEncoderConfig, VideoEncoderConfig
+from lerobot.configs import DEFAULT_DEPTH_UNIT, DepthEncoderConfig, VideoEncoderConfig
 from lerobot.utils.constants import HF_LEROBOT_HUB_CACHE
 
 from .dataset_metadata import CODEBASE_VERSION, LeRobotDatasetMetadata
@@ -58,7 +58,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
         download_videos: bool = True,
         video_backend: str | None = None,
         return_uint8: bool = False,
-        depth_output_unit: str = "mm",
+        depth_output_unit: str = DEFAULT_DEPTH_UNIT,
         batch_encoding_size: int = 1,
         camera_encoder: VideoEncoderConfig | None = None,
         depth_encoder: DepthEncoderConfig | None = None,
@@ -725,7 +725,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
         obj.episodes = None
         obj._video_backend = video_backend if video_backend is not None else get_safe_default_video_backend()
         obj._return_uint8 = False
-        obj._depth_output_unit = "mm"
+        obj._depth_output_unit = DEFAULT_DEPTH_UNIT
         obj._batch_encoding_size = batch_encoding_size
         obj._encoder_threads = encoder_threads
 
@@ -824,7 +824,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
         obj.episodes = None
         obj._video_backend = video_backend if video_backend else get_safe_default_video_backend()
         obj._return_uint8 = False
-        obj._depth_output_unit = "mm"
+        obj._depth_output_unit = DEFAULT_DEPTH_UNIT
         obj._batch_encoding_size = batch_encoding_size
 
         if obj._requested_root is not None: