wrappers
DefaultWrapper
Bases: EnvironmentWrapper
Default eval wrapper: low-resolution (224x224) RGB observations.
Camera resolution and modalities are declared via :meth:camera_spec and baked into the robot
config at env CREATION by the BatchedEvaluator
(see omnigibson.eval.evaluator.BatchedEvaluator.load_env),
not changed at runtime. This is required because in multi-env mode robot cameras are batched into
a single TiledVisionSensor whose resolution is fixed at creation, so a post-hoc per-sensor
resize would be silently ignored.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
env
|
Environment
|
The environment to wrap. |
required |
Source code in OmniGibson/omnigibson/eval/wrappers/default_wrapper.py
camera_spec()
classmethod
Returns {"modalities": [...], "resolution": {camera_role: (H, W)}} for this eval profile.
Source code in OmniGibson/omnigibson/eval/wrappers/default_wrapper.py
RGBDFullResWrapper
Bases: EnvironmentWrapper
Eval wrapper: full-resolution RGB-D observations (head at HEAD_RESOLUTION, wrists at WRIST_RESOLUTION) matching the data-collection cameras, plus a depth modality.
As with :class:~omnigibson.eval.wrappers.default_wrapper.DefaultWrapper, camera resolution and
modalities are declared via :meth:camera_spec and baked into the robot config at env CREATION
(multi-env tiled rendering fixes resolution at creation, so a runtime resize would not apply).
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
env
|
Environment
|
The environment to wrap. |
required |
Source code in OmniGibson/omnigibson/eval/wrappers/rgbd_full_res_wrapper.py
camera_spec()
classmethod
Returns {"modalities": [...], "resolution": {camera_role: (H, W)}} for this eval profile.