class TestTypeDataset:
def __init__(self, opt, use_difficult=True, id_file=None, img_dir=None, anno_dir=None):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir, id_file=id_file, img_dir=img_dir, anno_dir=anno_dir, use_difficult=use_difficult)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img = preprocess(ori_img)
return img, ori_img.shape[1:], bbox, label, difficult
def __len__(self):
return len(self.db)
class TestDataset:
def __init__(self, opt, split='test', use_difficult=True):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir, split=split, use_difficult=use_difficult)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img = preprocess(ori_img)
return img, ori_img.shape[1:], bbox, label, difficult
def __len__(self):
return len(self.db)
class TestDataset:
def __init__(self, opt):
self.opt = opt
self.db = VOCBboxDataset(self.opt.test_data_dir)
def __getitem__(self, idx):
ori_img, bbox, label, difficult, mask = self.db.get_example(idx)
img, mask = preprocess(ori_img, mask)
return img, ori_img.shape[1:], bbox, label, difficult, mask
def __len__(self):
return len(self.db)
class TestDataset(object):
def __init__(self):
self.db = VOCBboxDataset(
'D:/pyworks/FasterRCNN/data/testset/VOCdevkit/VOC2007', 'test',
True)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img = preprocess(ori_img)
return img, ori_img.shape[1:], bbox, label, difficult
def __len__(self):
return len(self.db)
class TrainDataset(object):
def __init__(self):
self.db = VOCBboxDataset(
'D:/pyworks/FasterRCNN/data/VOCdevkit/VOC2007', 'trainval')
self.transform = Transform()
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img, bbox, label, scale = self.transform((ori_img, bbox, label))
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
class Dataset:
def __init__(self, opt):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir)
self.tsf = Transform(opt.min_size, opt.max_size)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
class TestFGSMDataset:
def __init__(self, opt, split='test', use_difficult=True):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir, split=split, use_difficult=use_difficult)
self.tsf = Transform(opt.min_size, opt.max_size)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
_, _, _, scale = self.tsf((ori_img, bbox, label))
img = preprocess(ori_img)
return img, ori_img.shape[1:], bbox, label, difficult, scale
def __len__(self):
return len(self.db)
class TestDataset:
def __init__(self, configurations, split='val', use_difficult=True):
self.configurations = configurations
self.db = VOCBboxDataset(configurations.voc_data_dir,
split=split,
use_difficult=use_difficult)
def __getitem__(self, idx):
ori_image, bbox, label, difficult = self.db.get_example(idx)
image = preprocess(ori_image)
return image, ori_image.shape[1:], bbox, label, difficult
def __len__(self):
return len(self.db)
class Dataset:
def __init__(self, opt):
self.opt = opt
self.dataset = VOCBboxDataset(opt.voc_data_dir)
self.transform = Transform(opt.min_size, opt.max_size)
def __getitem__(self, item):
ori_img, bbox, label, difficult = self.dataset.get_example(item)
img, bbox, label, scale = self.transform((ori_img, bbox, label))
return img.copy(), bbox.copy(), label.copy(
), scale #返回处理后的图片(CHW,[-1,1]),bbox,label,原图和调整后图的比例因子
def __len__(self):
return len(self.dataset)
class Dataset:
def __init__(self):
self.db = VOCBboxDataset(voc_data_dir)
self.tsf = Transform(min_size=600, max_size=1000)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
class TestDataset:
def __init__(self, opt, split='test', use_difficult=True):
self.opt = opt
self.db = VOCBboxDataset('/data2/dechunwang/dataset/',
split=split,
use_difficult=use_difficult)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
# print('\n', '*' * 70, 'ori_img shape: ', ori_img.shape, '\n', '*' * 70)
img = preprocess(ori_img)
# print('\n', '*' * 70, 'img shape: ', img.shape, '\n', '*' * 70)
return ori_img, img, ori_img.shape[1:], bbox, label, difficult
def __len__(self):
return len(self.db)
class Dataset:
def __init__(self, opt):
self.opt = opt
self.db = VOCBboxDataset(opt.train_data_dir)
self.tsf = Transform()
def __getitem__(self, idx):
ori_img, bbox, label, difficult, mask = self.db.get_example(idx)
img, bbox, label, scale, mask = self.tsf((ori_img, bbox, label, mask))
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale, mask.copy()
def __len__(self):
return len(self.db)
class SmallImageDataset:
def __init__(self, opt):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir, split='trainval_small_32', small_obj=True)
self.tsf = Transform(opt.min_size, opt.max_size)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
class DatasetAugmented:
def __init__(self, opt, id_file=None, img_dir=None, anno_dir=None, use_diff=True):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir, id_file=id_file, img_dir=img_dir, anno_dir=anno_dir, use_difficult=use_diff)
self.tsf = Transform(opt.min_size, opt.max_size)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
class Dataset:
def __init__(self, opt):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir, opt.label_names)
self.tsf = Transform(opt.min_size, opt.max_size)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), ori_img.shape[1:], bbox.copy(), \
label.copy(), scale, difficult
def __len__(self):
return len(self.db)
class TestDataset:
# 测试集 有五个输出
def __init__(self, opt, split=opt.testtxt, use_difficult=True):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir,
split=split,
use_difficult=use_difficult)
def __getitem__(self, idx):
ori_img, bbox, label, difficult, id_ = self.db.get_example(idx)
if len(bbox) == 0:
return 0, ori_img.shape[1:], bbox, 0, 0, id_
img = preprocess(ori_img)
return img, ori_img.shape[1:], bbox, label, difficult, id_
def __len__(self):
return len(self.db)
class Dataset:
def __init__(self, opt):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir) # 实例化类
self.tsf = Transform(opt.min_size, opt.max_size) # 获得变换类
def __getitem__(self, idx): # 运行Dataset自动运行
# 调用VOCBboxDataset中的get_example()将img, bbox, label, difficult一个个取出
ori_img, bbox, label, difficult = self.db.get_example(idx)
# 调用前面的Transform函数将图片, label进行最小值最大值放缩归一化
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
class Dataset:
def __init__(self, opt):
self.opt = opt #预设参数
self.db = VOCBboxDataset(opt.voc_data_dir) #用预设参数创建数据集类
self.tsf = Transform(opt.min_size, opt.max_size) #实例化预处理类
def __getitem__(self, idx): #获得idx下标的一个数据
ori_img, bbox, label, difficult = self.db.get_example(
idx) #从数据集中获得一个batch的数据
img, bbox, label, scale = self.tsf((ori_img, bbox, label)) #预处理数据
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self): #获得数据集总长度
return len(self.db)
class Dataset:
def __init__(self, opt):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir)
self.tsf = Transform(opt.min_size, opt.max_size)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(
idx
) #use voc_dataset.py VOCBboxDataset get_example() return img, bbox, label, difficult
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db) #trainval =5011 test=4952
class Dataset: #训练集样本的生成
def __init__(self, opt):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir) #实例化 类
self.tsf = Transform(opt.min_size, opt.max_size)
def __getitem__(self, idx): #__ xxx__运行Dataset类时自动运行
ori_img, bbox, label, difficult = self.db.get_example(idx)
#调用VOCBboxDataset中的get_example（）从数据集存储路径中将img, bbox, label, difficult 一个个的获取出来
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
#调用前面的Transform函数将图片,label进行最小值最大值放缩归一化，重新调整bboxes的大小，然后随机反转，最后将数据集返回
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
class TestDataset:
def __init__(self, opt, split='test', use_difficult=True):
self.opt = opt
if self.opt.data == 'markers':
self.db = MarkersDataset(opt.data_dir)
elif self.opt.data == 'voc':
self.db = VOCBboxDataset(opt.data_dir)
else:
raise Exception('database type not recognised: {}'.format(
self.opt.data))
self.label_names = self.db.get_label_names()
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img = preprocess(ori_img)
return img, ori_img.shape[1:], bbox, label, difficult
def __len__(self):
return len(self.db)
class TestDataset:
def __init__(self, opt, split='test', use_difficult=True):
self.opt = opt
if opt.data == 'cityperson':
self.db = CityPersonTestset(
img_path=opt.cityperson_data_dir + '/leftImg8bit/val',
annotation_path=opt.cityperson_data_dir +
'/gtBboxCityPersons/val')
else:
self.db = VOCBboxDataset(opt.voc_data_dir,
split=split,
use_difficult=use_difficult)
def __getitem__(self, idx):
ori_img, bbox, label, difficult, image_id = self.db.get_example(idx)
img = preprocess(ori_img)
return img, ori_img.shape[1:], bbox, label, difficult, image_id
def __len__(self):
return len(self.db)
class Dataset:
def __init__(self, opt):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir)
self.tsf = Transform(opt.min_size, opt.max_size)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
# print("dataset getitem input: ", bbox, bbox.shape)
# print("difficult", type(difficult), difficult, difficult.shape)
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
#print(bbox, label)
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
#print("Dataset copied output:", bbox.copy(), bbox.copy().shape)
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
class VOC2007Detect(Dataset):
def __init__(self, path, is_train=True, min_size=600, max_size=1000):
self.path = path
self.is_train = is_train
if self.is_train:
self.db = VOCBboxDataset(path)
self.tsf = Transform(min_size, max_size)
else:
self.db = VOCBboxDataset(path, split='val', use_difficult=True)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
if self.is_train:
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
return img, bbox, label, scale
else:
img = preprocess(ori_img)
return img, ori_img.shape[1:], bbox, label, difficult
def __len__(self):
return len(self.db)
class Dataset:
def __init__(self, opt):
self.opt = opt
if opt.data == 'cityperson':
self.db = CityPersonTrainset(
img_path=opt.cityperson_data_dir + '/leftImg8bit/train',
annotation_path=opt.cityperson_data_dir +
'/gtBboxCityPersons/train')
else:
self.db = VOCBboxDataset(opt.voc_data_dir)
self.tsf = Transform(opt.min_size, opt.max_size)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
class Dataset:
# opt.voc_data_dir = '/home/featurize/data/VOCdevkit/VOC2007/'
# min_size = 600 # image resize
# max_size = 1000 # image resize
# min_size 与 max_size 主要是因为我们图片要求 短边<=600 ,长边<= 1000
def __init__(self, opt):
self.opt = opt
self.db = VOCBboxDataset(opt.voc_data_dir)
self.tsf = Transform(opt.min_size, opt.max_size)
# —getitem__可以简单的理解为从数据集存储路径中将例子一个个的获取出来，
# 然后调用前面的Transform函数将图片,label进行最小值最大值放缩归一化，
# 重新调整bboxes的大小，然后随机反转，最后将数据集返回！
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
class TrainDataset:
"""
训练使用的数据集
__getitem__ : 返回图片，包围盒，标签以及归一化图片时的放缩倍数
img :
shape = (channel, height/y-axis, width/x-axis)
dtype = float32
归一化处理后的图片，归一方法根据pytorch还是caffe不一样
bbox :
shape = (num_bbox, 4) => (y_{min}, x_{min}, y_{max}, x_{max})
dtype = float32
label :
shape = (num_bbox, )
dtype = int32
scale :
使用Transform将图片放缩到指定范围，bbox也随之发生改变
"""
def __init__(self, opt):
self.opt = opt
if opt.data == 'voc':
self.db = VOCBboxDataset(opt.voc_data_dir)
elif opt.data == 's3d':
self.db = S3DBboxDataset(opt.s3d_data_dir)
self.tsf = Transform(opt.min_size, opt.max_size)
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
class Dataset:
def __init__(self, opt):
self.opt = opt
if self.opt.data == 'markers':
self.db = MarkersDataset(opt.data_dir)
elif self.opt.data == 'voc':
self.db = VOCBboxDataset(opt.data_dir)
else:
raise Exception('database type not recognised: {}'.format(
self.opt.data))
self.tsf = Transform(opt.min_size, opt.max_size)
self.label_names = self.db.get_label_names()
def __getitem__(self, idx):
ori_img, bbox, label, difficult = self.db.get_example(idx)
img, bbox, label, scale = self.tsf((ori_img, bbox, label))
# TODO: check whose stride is negative to fix this instead copy all
# some of the strides of a given numpy array are negative.
return img.copy(), bbox.copy(), label.copy(), scale
def __len__(self):
return len(self.db)
