MCPcopy Create free account
hub / github.com/bbaaii/DreamDiffusion / augmentation

Function augmentation

code/dataset.py:55–70  ·  view source on GitHub ↗

data: num_samples, num_voxels_padded return: data_aug: num_samples*aug_times, num_voxels_padded

(data, aug_times=2, interpolation_ratio=0.5)

Source from the content-addressed store, hash-verified

53 return v_split
54
55def augmentation(data, aug_times=2, interpolation_ratio=0.5):
56 '''
57 data: num_samples, num_voxels_padded
58 return: data_aug: num_samples*aug_times, num_voxels_padded
59 '''
60 num_to_generate = int((aug_times-1)*len(data))
61 if num_to_generate == 0:
62 return data
63 pairs_idx = np.random.choice(len(data), size=(num_to_generate, 2), replace=True)
64 data_aug = []
65 for i in pairs_idx:
66 z = interpolate_voxels(data[i[0]], data[i[1]], interpolation_ratio)
67 data_aug.append(np.expand_dims(z,axis=0))
68 data_aug = np.concatenate(data_aug, axis=0)
69
70 return np.concatenate([data, data_aug], axis=0)
71
72def interpolate_voxels(x, y, ratio=0.5):
73 ''''

Callers

nothing calls this directly

Calls 1

interpolate_voxelsFunction · 0.85

Tested by

no test coverage detected