Dark

class plugins.extract.align.dark_decoder.Dark(num_points: int, size: int, blur_kernel: int = 11)

Bases: object

Dark heatmap decoding

https://github.com/ilovepose/DarkPose

Parameters:

  • num_points (int) – The number of landmarks output from the model

  • size (int) – The size of the heatmap

  • blur_kernel (int)

Methods Summary

__call__(heatmap)

Call self as a function.

gaussian_blur(heatmap)

Perform gaussian blurring on the heatmaps

get_max_preds(batch_heatmaps)

get predictions from score maps

taylor(heatmap, coords)

Sub-pixel refine the predictions

Methods Documentation

__call__(heatmap: ndarray)

Call self as a function.

Parameters:

heatmap (ndarray)

gaussian_blur(heatmap: ndarray) → ndarray

Perform gaussian blurring on the heatmaps

Parameters:

heatmap (ndarray) – Batch of heatmaps to blur (N, points, size, size)

Return type:

The blurred heatmaps

get_max_preds(batch_heatmaps: ndarray) → ndarray

get predictions from score maps

Parameters:

  • heatmaps – Heatmap to derive points from ([batch_size, num_joints, height, width])

  • batch_heatmaps (ndarray)

Returns:

The derived points from the heatmaps (B, N, 2)

Return type:

coords

taylor(heatmap: ndarray, coords: ndarray) → ndarray

Sub-pixel refine the predictions

Parameters:

  • heatmap (ndarray) – The processed heatmaps for refinement

  • coords (ndarray) – The coordinates to be refined

Return type:

The refined coordinates

__call__(heatmap: ndarray)

Call self as a function.

Parameters:

heatmap (ndarray)

gaussian_blur(heatmap: ndarray) → ndarray

Perform gaussian blurring on the heatmaps

Parameters:

heatmap (ndarray) – Batch of heatmaps to blur (N, points, size, size)

Return type:

The blurred heatmaps

get_max_preds(batch_heatmaps: ndarray) → ndarray

get predictions from score maps

Parameters:

  • heatmaps – Heatmap to derive points from ([batch_size, num_joints, height, width])

  • batch_heatmaps (ndarray)

Returns:

The derived points from the heatmaps (B, N, 2)

Return type:

coords

taylor(heatmap: ndarray, coords: ndarray) → ndarray

Sub-pixel refine the predictions

Parameters:

  • heatmap (ndarray) – The processed heatmaps for refinement

  • coords (ndarray) – The coordinates to be refined

Return type:

The refined coordinates