DPVO

class grid.model.perception.vo.vo_dpvo.DPVO(*args, **kwargs)

DPVO: Visual Odometry Model

This class implements a wrapper for the DPVO model, which estimates camera poses from RGB images using a deep learning approach.

Credits:

https://github.com/princeton-vl/DPVO

License:

This code is licensed under the MIT License.

__init__(calib=array([320, 320, 320, 240]))

Return type:

None

run(image)

Uses DPVO to predict the camera pose for the given RGB image relative to the previous one. If this is the first image, initializes the pose estimation routine.

Parameters:

image (np.ndarray) -- The input RGB image.

Returns:

The predicted pose as a 1x6 tensor containing X, Y, Z positions and R, P, Y orientation.

Return type:

torch.Tensor

Example

>>> from grid.model.perception.vo.vo_dpvo import DPVO
>>> calib = np.array([320, 320, 320, 240])
>>> vo = DPVO(calib)
>>> pose = vo.run(img)
>>> print(pose)

terminate(reconstruct=False)

Terminates the prediction routine and clears the keyframes.

Parameters:

reconstruct (bool) -- If True, display the reconstructed sparse map as a point cloud in Rerun.

Returns:

A tuple containing the predicted camera poses and the reconstructed sparse map.

Return type:

Tuple[np.ndarray, np.ndarray]