miop.raft_flow

Classes

RaftFlow([device])

Computes dense optical flow (displacement maps) between image pairs using the RAFT model.

class miop.raft_flow.RaftFlow(device=None)

Bases: DAGNode

Computes dense optical flow (displacement maps) between image pairs using the RAFT model.

device

The device (CPU, CUDA, MPS) used for inference.

Type:

torch.device

disp_maps

The computed displacement maps for each image pair in each face of the collection. Each displacement map is a 2D flow field of shape (2, H, W), where:

• disp[0, :, :] is the x-direction displacement,

• disp[1, :, :] is the y-direction displacement.

Type:

list of list of np.ndarray or None

compute_flow(image_1, image_2)

Computes the dense optical flow (displacement map) between two RGB images using the RAFT model.

Parameters:

• image_1 (np.ndarray) – The first image of shape (H, W, 3) in uint8 or float format.

• image_2 (np.ndarray) – The second image of shape (H, W, 3) in uint8 or float format.

Returns:

A displacement map of shape (2, H, W), where the first channel is x-displacement and the second channel is y-displacement.

Return type:

np.ndarray

eval(img_collection)

Evaluates the optical flow for each image pair in the image collection.

Parameters:

img_collection (ImageCollection) – An image collection containing pairs of images organized by faces.

Returns:

A nested list where each element corresponds to a face, and contains the displacement maps for each image pair in that face.

Return type:

list of list of np.ndarray

show(face=0, pair=0, cmap='Spectral')

Visualizes the displacement map for a specific image pair using matplotlib.

Parameters:

• face (int, optional) – Index of the face to visualize (default: 0).

• pair (int, optional) – Index of the image pair within the face to visualize (default: 0).

• cmap (str, optional) – Colormap to use for the visualization (default: “Spectral”).

Raises:

AttributeError – If self.disp_maps is not available.