Resize

class torchrl.envs.transforms.Resize(w: int, h: int | None = None, interpolation: str = 'bilinear', in_keys: Sequence[NestedKey] | None = None, out_keys: Sequence[NestedKey] | None = None)

Resizes a pixel observation.

Parameters:

• w (int) – resulting width.

• h (int, optional) – resulting height. If not provided, the value of w is taken.

• interpolation (str) – interpolation method

Examples

>>> from torchrl.envs import GymEnv
>>> t = Resize(64, 84)
>>> base_env = GymEnv("HalfCheetah-v4", from_pixels=True)
>>> env = TransformedEnv(base_env, Compose(ToTensorImage(), t))

transform_observation_spec(observation_spec: TensorSpec) → TensorSpec

Transforms the observation spec such that the resulting spec matches transform mapping.

Parameters:

observation_spec (TensorSpec) – spec before the transform

Returns:

expected spec after the transform