nicetoolbox.detectors.feature_detectors.gaze_multiview.gaze_fusion.GazeFusion

class nicetoolbox.detectors.feature_detectors.gaze_multiview.gaze_fusion.GazeFusion(io, data, sequence_context, algorithm_instance: str)

Bases: BaseFeature

Computes a single 3D gaze vector from multiple camera views and/or multiple algorithms.

Given the fused 3D gaze vectors, the user can optionally select to apply a Savitzky-Golay filter for temporal smoothing. Finally, the fused 3D gaze vectors are projected back to each camera view to obtain 2D gaze points for visualization (If enabled).

Expected Input:

• Method detectors outputting ‘3d’ array.

• Shape: (Subjects, Cameras, Frames, 3)

• Optional: ‘confidence_scores’ (Subjects, Cameras, Frames) for weighted fusion.

Initialize base detector with references.

Subclasses should call super().__init__() and set inference_config.

Methods

compute	Run the fusion pipeline.
compute_output_folders	Compute extra output folders for all components.
compute_result_folders	Compute result folders for all components.
compute_viz_folders	Compute visualization folders for all components.
get_input_file	Get the input file path for a specific upstream detector.
run	Execute feature detector: compute() + post_compute().
visualization	Generates visualization images with fused gaze overlays.

Attributes

algorithm_type
components
predictions_mapping	Access predictions mapping from runtime config.
requires_out_folder
data
io
sequence_context
detector_config
algorithm_instance
inference_config
visualize

compute() → Dict[str, Any] | None

Run the fusion pipeline.

Returns:

A list of dictionaries containing fused results.

compute_output_folders(requires_out_folder: bool) → Dict[str, str]

Compute extra output folders for all components.

compute_result_folders() → Dict[str, str]

Compute result folders for all components.

compute_viz_folders(visualize: bool) → Dict[str, str]

Compute visualization folders for all components.

get_input_file(component: str, algorithm: str) → Path

Get the input file path for a specific upstream detector.

Parameters:

• component – Component name (e.g., ‘body_joints’)

• algorithm – Algorithm name (e.g., ‘hrnetw48’)

Returns:

Path to the .npz result file

property predictions_mapping

Access predictions mapping from runtime config.

run() → Any

Execute feature detector: compute() + post_compute().

Returns computed data for visualization.

visualization(result: Dict[str, Any]) → None

Generates visualization images with fused gaze overlays.