draft: Organize ros params as dataclass

yolov8_ros/parameters.py

@@ -0,0 +1,44 @@
+from dataclasses import dataclass, field
+
+from rclpy.node import Node
+
+
+@dataclass
+class YoloSkeletonRightWristNodeParams:
+    rgb_sub_topic: str = field(default="/camera/color/image_raw/compressed")
+    """sensor_msgs/CompressedImage RGB topic to subscribe to."""
+
+    depth_sub_topic: str = field(default="/camera/aligned/depth_to_color/image_raw")
+    """
+    sensor_msgs/Image aligned depth topic to subscribe to.
+
+    Images published to this topic must share the same intrinsic matrix as the
+    RGB image topic.
+
+    The uncompressed topic is used as there seems to be some difficulty
+    subscribing or decoding the compressed depth topic.
+    """
+
+    camera_info_sub_topic: str = field(default="/camera/color/camera_info")
+    """sensor_msgs/CameraInfo topic to subscribe to."""
+
+    stretch_robot_rotate_image_90deg: bool = field(default=True)
+    """Whether to rotate the input images 90 degrees clockwise before inference.
+
+    This parameter exists because the Stretch robot's camera is mounted
+    sideways, and YOLOv8 does not appear to perform well on rotated images.
+
+    TODO: This could be implemented better, but it works for our use case at the
+    time of writing.
+    """
+
+    @staticmethod
+    def read_from_ros(node: Node) -> "YoloSkeletonRightWristNodeParams":
+        params = YoloSkeletonRightWristNodeParams()
+
+        for param_name, default_value in vars(params).items():
+            ros_value = node.declare_parameter(param_name, default_value).value
+            assert type(ros_value) == type(default_value)
+            setattr(params, param_name, ros_value)
+
+        return params