Orthogonality-Explainability

Tests whether orthogonal convolutional channels represent more independent visual concepts, with scoring and labels for attribution.