These docs describe unreleased changes on main. See the latest release.

rlmesh.recorder.MediaRef

classfrom rlmesh.recorder import MediaRef[source]

A per-episode media asset referenced by a bundle-relative path.

MediaRef(
    camera: str,
    kind: str,
    path: str,
    format: str,
    frame_count: int | None = None,
    width: int | None = None,
    height: int | None = None,
    fps: float | None = None,
)

kind="video" is a video file living in the bundle at path – either recorded from the env’s image observations (AV1 mp4, encoded in process) or an env-produced container carried verbatim. frame_count/width/height/fps are filled for recorded videos. This is just the manifest row; the bytes are in the bundle.

camera

attribute[source]
camera: str

kind

attribute[source]
kind: str

path

attribute[source]
path: str

format

attribute[source]
format: str

frame_count

attribute[source]
frame_count: int | None

width

attribute[source]
width: int | None

height

attribute[source]
height: int | None

fps

attribute[source]
fps: float | None

to_dict

function[source]
to_dict() -> dict[str, Any]

JSON-native manifest row for this asset.