refactor(cameras): add warm-up, fix defaul args, remove width and height from find_cameras utils

lerobot/common/cameras/configs.py

@@ -33,7 +33,7 @@ class Cv2Rotation(Enum):
     ROTATE_270 = -90
 
 
-@dataclass
+@dataclass(kw_only=True)
 class CameraConfig(draccus.ChoiceRegistry, abc.ABC):
     fps: int | None = None
     width: int | None = None

lerobot/common/cameras/intel/camera_realsense.py

@@ -336,6 +336,10 @@ class RealSenseCamera(Camera):
 
         logger.debug(f"Validating stream configuration for {self.serial_number}...")
         self._validate_capture_settings()
+
+        logger.debug(f"Reading a warm-up frame for {self.serial_number}...")
+        self.read()  # NOTE(Steven): For now we just read one frame, we could also loop for X secs
+
         logger.info(f"Camera {self.serial_number} connected and configured successfully.")
 
     def _validate_fps(self, stream) -> None:
@@ -396,7 +400,7 @@ class RealSenseCamera(Camera):
 
     def read(
         self, color_mode: ColorMode | None = None, timeout_ms: int = 5000
-    ) -> Union[np.ndarray, Tuple[np.ndarray, np.ndarray]]:
+    ) -> np.ndarray | Tuple[np.ndarray, np.ndarray]:
         """
         Reads a single frame (color and optionally depth) synchronously from the camera.
 

lerobot/common/cameras/opencv/camera_opencv.py

@@ -205,6 +205,10 @@ class OpenCVCamera(Camera):
 
         logger.debug(f"Successfully opened camera {self.index_or_path}. Applying configuration...")
         self._configure_capture_settings()
+
+        logger.debug(f"Reading a warm-up frame for {self.serial_number}...")
+        self.read()  # NOTE(Steven): For now we just read one frame, we could also loop for X secs\
+
         logger.debug(f"Camera {self.index_or_path} connected and configured successfully.")
 
     def _validate_fps(self) -> None:

lerobot/common/cameras/opencv/configuration_opencv.py

@@ -1,4 +1,4 @@
-from dataclasses import dataclass, field
+from dataclasses import dataclass
 from pathlib import Path
 
 from ..configs import CameraConfig, ColorMode, Cv2Rotation
@@ -18,9 +18,7 @@ class OpenCVCameraConfig(CameraConfig):
     ```
     """
 
-    index_or_path: int | Path = field(
-        default=...,
-    )
+    index_or_path: int | Path
     color_mode: ColorMode = ColorMode.RGB
     channels: int = 3  # NOTE(Steven): Why is this a config?
     rotation: Cv2Rotation = Cv2Rotation.NO_ROTATION

lerobot/find_cameras.py

@@ -157,28 +157,21 @@ def create_camera_instance(cam_meta: Dict[str, Any]) -> Optional[Dict[str, Any]]
     """Create and connect to a camera instance based on metadata."""
     cam_type = cam_meta.get("type")
     cam_id = cam_meta.get("id")
-    default_profile = cam_meta.get("default_stream_profile")
-    width = default_profile.get("width")
-    height = default_profile.get("height")
     instance = None
 
-    logger.info(f"Preparing {cam_type} ID {cam_id} with profile: Width={width}, Height={height}")
+    logger.info(f"Preparing {cam_type} ID {cam_id} with default profile")
 
     try:
         if cam_type == "OpenCV":
             cv_config = OpenCVCameraConfig(
                 index_or_path=cam_id,
                 color_mode=ColorMode.RGB,
-                width=width,
-                height=height,
             )
             instance = OpenCVCamera(cv_config)
         elif cam_type == "RealSense":
             rs_config = RealSenseCameraConfig(
                 serial_number=str(cam_id),
                 color_mode=ColorMode.RGB,
-                width=width,
-                height=height,
             )
             instance = RealSenseCamera(rs_config)
         else:
@@ -270,6 +263,7 @@ def save_images_from_all_cameras(
     logger.info(f"Starting image capture for {record_time_s} seconds from {len(cameras_to_use)} cameras.")
     start_time = time.perf_counter()
 
+    # NOTE(Steven): This seems like an overkill to me
     with concurrent.futures.ThreadPoolExecutor(max_workers=len(cameras_to_use) * 2) as executor:
         try:
             while time.perf_counter() - start_time < record_time_s: