MIR.models.LessNet

class MIR.models.LessNet(*args, **kwargs)[source]

MIR-callable LessNet registration model.

The upstream normalized-grid output is converted to voxel units and, if SVF is enabled, integrated by scaling and squaring.

Parameters:

  • configs – Config object with LessNet hyperparameters (defaults to get_LessNet_default_config()).

  • gen_output – If True, also return the warped moving image.

  • SVF – Optional override for stationary velocity field integration.

  • SVF_steps – Optional override for scaling-and-squaring steps.

Inputs:

input_imgs – Tuple (mov, fix) of tensors [B, C, D, H, W].

Returns:

MIR-convention voxel displacement field [B, 3, D, H, W], or (warped, flow) when gen_output is True.

__init__(configs=None, gen_output=False, SVF=None, SVF_steps=None)[source]

Initialize MIR LessNet.

Parameters:

  • configs – Config object with LessNet hyperparameters.

  • gen_output – If True, also return the warped moving image.

  • SVF – Optional override for stationary velocity field integration.

  • SVF_steps – Optional override for scaling-and-squaring steps.

Methods

__init__([configs, gen_output, SVF, SVF_steps])

Initialize MIR LessNet.

forward(input_imgs)

Run LessNet registration.