quantize::laplacian::LaplacianWeightedError¶

Public Functions¶

	Name
	init(self self, float sigma =2.0, str mode ="reflect")
float	compute(self self, np.ndarray original_2d, np.ndarray reconstructed_2d, int block_size)

Protected Attributes¶

	Name
	_sigma
	_mode

Detailed Description¶

class quantize::laplacian::LaplacianWeightedError;

Compute Laplacian pyramid-weighted error for encode-side block selection.

Constructor kwargs (documented per RESEARCH.md Pattern 2 for tunability):
    sigma: Base sigma for Gaussian smoothing per level. Actual sigma per
           level is sigma * 2**level. Default: 2.0.
    mode:   Boundary handling mode passed to scipy.ndimage.gaussian_filter.
            Default: 'reflect'.

Public Functions Documentation¶

function init¶

__init__(
    self self,
    float sigma =2.0,
    str mode ="reflect"
)

function compute¶

float compute(
    self self,
    np.ndarray original_2d,
    np.ndarray reconstructed_2d,
    int block_size
)

Compute Laplacian-weighted MSE between original and reconstructed.

Args:
    original_2d: 2D numpy array of original float32 weights.
    reconstructed_2d: 2D numpy array of quantized+dequantized weights.
    block_size: Edge size of the block (4, 8, 16, 32, or 64).

Returns:
    float: Laplacian-weighted MSE. For blocks <= 8x8, returns plain
           MSE (Laplacian skipped per Pitfall 2).

Protected Attributes Documentation¶

variable _sigma¶

_sigma;

variable _mode¶

_mode;

Updated on 2026-07-25 at 22:56:57 +0000