1 vin.modules.semidense_grid

vin.modules.semidense_grid

Semidense projection-grid encoders for VIN scorers.

This module owns reusable torch.nn.Module blocks that transform per-candidate screen-space semidense projection grids into compact feature vectors. The geometry helper aria_nbv.vin.geometry.semidense_projection.build_projection_grid constructs the deterministic grid; the encoder defined here owns the learnable convolutional projection used by the active VIN v3 scorer.

1.1 Classes

Name Description
SemidenseGridEncoder Encode semidense projection grids with a small convolutional head.
SemidenseGridEncoderConfig Configure the VIN semidense projection-grid CNN.

1.1.1 SemidenseGridEncoder

vin.modules.semidense_grid.SemidenseGridEncoder(config)

Encode semidense projection grids with a small convolutional head.

The class subclasses torch.nn.Sequential intentionally: when assigned to VinModelV3.semidense_cnn, learnable parameters keep the historical checkpoint keys semidense_cnn.0.*, semidense_cnn.2.*, and semidense_cnn.6.*. Do not wrap the layers in a child module unless a checkpoint migration is planned.

1.1.1.1 Methods

Name Description
encode_projection_features Return one CNN feature vector per batch/candidate projection.
1.1.1.1.1 encode_projection_features
vin.modules.semidense_grid.SemidenseGridEncoder.encode_projection_features(
    proj_data,
    *,
    batch_size,
    num_candidates,
    device,
    dtype,
)

Return one CNN feature vector per batch/candidate projection.

1.1.1.1.1.1 Parameters
Name Type Description Default
proj_data dict[str, Tensor] Projection output from aria_nbv.vin.geometry.semidense_projection.project_points_to_candidate_cameras. required
batch_size int Batch size B. required
num_candidates int Number of candidate cameras N_q. required
device torch.device Device for the intermediate projection grid. required
dtype torch.dtype Floating point dtype for the grid and returned features. required
1.1.1.1.1.2 Returns
Name Type Description
Tensor Tensor["B N_q F", float32] where F is
Tensor SemidenseGridEncoderConfig.out_dim.

1.1.2 SemidenseGridEncoderConfig

vin.modules.semidense_grid.SemidenseGridEncoderConfig()

Configure the VIN semidense projection-grid CNN.

1.1.2.1 Attributes

Name Description
target_type Factory target for BaseConfig.setup_target.
grid_size Spatial side length of the deterministic projection grid.
channels Hidden channel width for the convolutional layers.
out_dim Output feature dimension per candidate projection.
Back to top