ros-python-conversions

Conversions between ROS 2 messages and Python types in data-models.

Motivation

Few algorithms should depend on raw ROS message objects end-to-end. A dedicated conversion layer keeps bag-reading code small and pushes quirks (encodings, transports, stamps) into one place.

For chronological data, each converted value should carry at least timestamp and index via BaseMetadata on the corresponding BaseInstance.

Layout (examples)

Module	Purpose
ros2/raw_rgb_image.py	sensor_msgs/Image, CompressedImage → ImageInstance (BGR).
ros2/depth_image.py	Depth sensor_msgs/Image → ImageInstance (float32 (H, W); 16UC1 / mono16 scaled mm→m).
ros2/ffmpeg_transport.py	Detect FFMPEGPacket; concatenate/decode H.264/HEVC with PyAV (av), with Annex-B vs length-prefixed handling and optional ffmpeg CLI fallback.
ros2/time.py, tf.py, odometry.py	Stamps and common nav message conversions.

Implementing new conversions

1. Define or reuse a data-models type for the payload (fields + BaseMetadata).
2. Add functions under ros_python_conversions/ros2/, following existing modules: message → instance for bag replay; add instance → message only if you need publishing or round-trips.
3. For bag streams, use your decode function as the decode_fn passed into Ros2DataStream (see data-streams README).

Dependencies: depth and RGB paths use cv_bridge (ROS environment). FFMPEGPacket decoding requires av (PyAV); system ffmpeg is optional for difficult bitstreams when width/height are known.