crowdcount.data.data_preprocess

crowdcount.data.data_preprocess.adaptive_gaussian(gt, mode='kdtree')

Parameters

• gt (numpy.ndarray) – the ground truth to be processed by gaussian filter

• mode (str, optional) – the way to generate density map. “uniform”:

Returns

numpy.ndarray

crowdcount.data.data_preprocess.uniform_gaussian(gt, sigma=15, radius=4)

Parameters

• gt (numpy.ndarray) – the ground truth to be processed by gaussian filter

• sigma (scalar or sequence of scalars, optional) – Standard deviation for Gaussian kernel. The standard deviations of the Gaussian filter are given for each axis as a sequence, or as a single number, in which case it is equal for all axes.

• radius (int, optional) – the radius of gaussian kernel

Returns

numpy.ndarray

class crowdcount.data.data_preprocess.PreProcess(root, name='shtu_a')

generate density map

Parameters

• root (str) – the root of dataset, only support shtu_a, shtu_b, ucf_qnrf and ucf_cc_50 now. For ShanghaiTech part A and part B, the root should be upper dir over “part_A_final” or “part_B_final”. And for UCF QNRF and UCF CC 50, the root should be upper dir over “Train” and “Test”

• name (str, optional) – the name of dataset, must be in [“shtu_a”, “shtu_b”, “ucf_qnrf”, “ucf_cc”]. default: shtu_a

process(mode='uniform', sigma=4, radius=7)

generate density map and save.

Parameters

• mode (str, optional) – the way to generate gaussian filter. “uniform”: uniform sigma and radius of filter, the suggestion from C-3-Framework is sigma 15 and radius 7 as the default. “adaptive_kdtree”: use adaptive gaussian kernel with kdtree. “adaptive_voronio”: use adaptive gaussian kernel with voronio map. defalut: uniform.

• sigma (int, optional) – the sigma of gaussian filter. default: 4.

• radius (int, optional) – the radius of gaussian area. default: 7.