data-streams

Iterators and random access over chronological sensor data for offline analysis.

Design pattern

• DataStream (in core/data_stream.py) — Abstract chronological API: timestamps, get_instance, nearest-by-time queries, optional interpolation flags, etc.
• impl/ros2.py — Ros2DataStream — Backs a stream from a ROS 2 bag (directory or .mcap) and a single topic; deserializes with rosbags and calls a decode_fn(msg, index, timestamp) that returns a BaseInstance.
• impl/ros2_ffmpeg.py — Ros2FfmpegPacketStream — Same bag/topic wiring, but decodes ffmpeg_image_transport / FFMPEGPacket (e.g. H.264/HEVC) to BGR frames and returns ImageInstance by index.
• ros2_common/camera_streams.py — make_rgb_image_stream picks Ros2FfmpegPacketStream when the topic type is FFMPEGPacket, otherwise plain Ros2DataStream with RGB/compressed image decoding. make_depth_image_stream wires depth sensor_msgs/Image → float depth grids via ros-python-conversions.
• collection_streams/ — Higher-level streams that combine multiple bag topics (e.g. TF-derived poses).

Implementing a new source: subclass DataStream, supply ordered timestamps, and implement make_instance (and any metadata helpers your base class expects).