|
- # -*- coding: utf-8 -*-
- """
- @author: huangxs
- @License: (C)Copyright 2021, huangxs
- @CreateTime: 2021/11/16 19:10:00
- @Filename: data process
-
- """
- import numpy as np
- from collections import OrderedDict
- from src.utils.dataset import MoNuSegGenerator, MoNuSegPreparedGenerator
- from src.utils.direction_transform import get_transforms_list
-
- import mindspore.dataset as ds
-
-
- def data_processing():
- transform_dict = OrderedDict([('random_color', 1), ('horizontal_flip', True), ('vertical_flip', True),
- ('random_chooseAug', 1), ('random_crop', 256),
- ('label_encoding', [3, 2, 1])])
- transform_list = get_transforms_list(transform_dict)
-
- # 构建用于训练的dataset
- monuseg = MoNuSegGenerator(
- image_dir='data/MoNuSeg_oridata/images/train_300',
- weight_dir='data/MoNuSeg_oridata/weight_maps/train_300',
- label_dir='data/MoNuSeg_oridata/labels/train_300',
- transform_list=transform_list)
- dataset = ds.GeneratorDataset(monuseg, ['input', 'weight_map', 'target0', 'target_point0', 'target_direction0'])
- dataset = dataset.batch(1)
- dataset = dataset.repeat(1)
-
- # 测试构建的数据集迭代
- for i, data in enumerate(dataset.create_dict_iterator()):
- print('i=%d' % i, data['input'].shape, data['weight_map'].shape,
- data['target0'].shape, data['target_point0'].shape, data['target_direction0'].shape)
- np.savez('data_prepare/train/train_data_%d' % i, input=data['input'].asnumpy(),
- weight_map=data['weight_map'].asnumpy(), target0=data['target0'].asnumpy(),
- target_point0=data['target_point0'].asnumpy(), target_direction0=data['target_direction0'].asnumpy())
-
- print('finish data processing before train')
-
-
- if __name__ == "__main__":
- data_processing()
|