初始化DataLoader类时必须注入一个参数dataset,而dataset为自己定义。DataSet类可以继承,但是必须重载__len__()和__getitem__
使用Pytoch封装的DataLoader有以下好处:
①可以自动实现多进程加载
②自动惰性加载,不会占用过多内存
③封装有数据预处理和数据增强等操作,避免重复造轮子
以Faster R-CNN为例,一般建议至少传入以下参数,方便后续使用:
class FRCNNDataset(Dataset):
def __init__(self, annotation_lines, input_shape = [600, 600], train = True):
self.annotation_lines = annotation_lines #数据集列表
self.length = len(annotation_lines) #数据集大小
self.input_shape = input_shape #输出尺寸
self.train = train #是否训练
然后重载__len__()和__getitem__
def __len__(self):
return self.length #直接返回长度
def __getitem__(self, index):
index = index % self.length
#训练时候对数据进行随机增强,但验证时不进行
image, y = self.get_random_data(self.annotation_lines[index], self.input_shape[0:2], random = self.train)
#将图片转换成矩阵
image = np.transpose(preprocess_input(np.array(image, dtype=np.float32)), (2, 0, 1))
#编码先验框
box_data = np.zeros((len(y), 5))
if len(y) > 0:
box_data[:len(y)] = y
box = box_data[:, :4]
label = box_data[:, -1]
return image, box, label
关于数据增强函数get_random_data(),其中还包含了对图片的无变形缩放功能
def get_random_data(self, annotation_line, input_shape, jitter=.3, hue=.1, sat=0.7, val=0.4, random=True):
# 数据经过处理后格式为:地址——(空格)——预测框,使用split函数即可切割出地址和先验框
line = annotation_line.split()
# 读取图像并转换为RGB格式
image = Image.open(line[0])
image = cvtColor(image)
# 获得图像的高宽与目标高宽
iw, ih = image.size
h, w = input_shape
# 读取先验框
box = np.array([np.array(list(map(int,box.split(',')))) for box in line[1:]])
仅缩放的无变形缩放功(非训练模式)
# 在不进行随机数据增强的情况下(非训练模式),直接变形后输出
if not random:
#获取变形比例
scale = min(w/iw, h/ih)
nw = int(iw*scale)
nh = int(ih*scale)
dx = (w-nw)//2
dy = (h-nh)//2
# 将图像多余的部分加上灰条
image = image.resize((nw,nh), Image.BICUBIC)
new_image = Image.new('RGB', (w,h), (128,128,128))
new_image.paste(image, (dx, dy))
image_data = np.array(new_image, np.float32)
# 对真实框进行调整
if len(box)>0:
np.random.shuffle(box)
box[:, [0,2]] = box[:, [0,2]]*nw/iw + dx
box[:, [1,3]] = box[:, [1,3]]*nh/ih + dy
box[:, 0:2][box[:, 0:2]<0] = 0
box[:, 2][box[:, 2]>w] = w
box[:, 3][box[:, 3]>h] = h
box_w = box[:, 2] - box[:, 0]
box_h = box[:, 3] - box[:, 1]
box = box[np.logical_and(box_w>1, box_h>1)] # discard invalid box
#返回图片和先验框
return image_data, box
带数据增强的无变形缩放(训练模式)
# 对图像进行缩放并且进行长和宽的扭曲
new_ar = iw/ih * self.rand(1-jitter,1+jitter) / self.rand(1-jitter,1+jitter)
scale = self.rand(.25, 2)
if new_ar < 1:
nh = int(scale*h)
nw = int(nh*new_ar)
else:
nw = int(scale*w)
nh = int(nw/new_ar)
image = image.resize((nw,nh), Image.BICUBIC)
# 将图像多余的部分加上灰条
dx = int(self.rand(0, w-nw))
dy = int(self.rand(0, h-nh))
new_image = Image.new('RGB', (w,h), (128,128,128))
new_image.paste(image, (dx, dy))
image = new_image
# 翻转图像
flip = self.rand()<.5
if flip: image = image.transpose(Image.FLIP_LEFT_RIGHT)
image_data = np.array(image, np.uint8)
# 对图像进行色域变换
# 计算色域变换的参数
r = np.random.uniform(-1, 1, 3) * [hue, sat, val] + 1
# 将图像转到HSV上
hue, sat, val = cv2.split(cv2.cvtColor(image_data, cv2.COLOR_RGB2HSV))
dtype = image_data.dtype
# 应用变换
x = np.arange(0, 256, dtype=r.dtype)
lut_hue = ((x * r[0]) % 180).astype(dtype)
lut_sat = np.clip(x * r[1], 0, 255).astype(dtype)
lut_val = np.clip(x * r[2], 0, 255).astype(dtype)
image_data = cv2.merge((cv2.LUT(hue, lut_hue), cv2.LUT(sat, lut_sat), cv2.LUT(val, lut_val)))
image_data = cv2.cvtColor(image_data, cv2.COLOR_HSV2RGB)
# 对真实框进行调整
if len(box)>0:
np.random.shuffle(box)
box[:, [0,2]] = box[:, [0,2]]*nw/iw + dx
box[:, [1,3]] = box[:, [1,3]]*nh/ih + dy
if flip: box[:, [0,2]] = w - box[:, [2,0]]
box[:, 0:2][box[:, 0:2]<0] = 0
box[:, 2][box[:, 2]>w] = w
box[:, 3][box[:, 3]>h] = h
box_w = box[:, 2] - box[:, 0]
box_h = box[:, 3] - box[:, 1]
box = box[np.logical_and(box_w>1, box_h>1)]
return image_data, box
关于collate_fn参数
__getitem__一般返回(image,label)样本对,而DataLoader需要一个batch_size用于处理batch样本,以便于批量训练。
默认的default_collate(batch)函数仅能对尺寸一致且batch_size相同的image进行整理,如将(img0,lbl0),(img1,lbl1),(img2,lbl2)整合为([img0,img1,img2],[lbl0,lbl1,lbl2]),如图像中含有box等参数则需要自定义处理
def frcnn_dataset_collate(batch):
images = []
bboxes = []
labels = []
for img, box, label in batch:
images.append(img)
bboxes.append(box)
labels.append(label)
images = torch.from_numpy(np.array(images))
return images, bboxes, labels
①在__getitem__中不需要获取box值,转而获取标志图png。
def __getitem__(self, index):
annotation_line = self.annotation_lines[index]
name = annotation_line.split()[0]
# 从文件中读取图像
jpg = Image.open(os.path.join(os.path.join(self.dataset_path, "VOC2007/JPEGImages"), name + ".jpg"))
png = Image.open(os.path.join(os.path.join(self.dataset_path, "VOC2007/SegmentationClass"), name + ".png"))
# 数据增强
jpg, png = self.get_random_data(jpg, png, self.input_shape, random = self.train)
jpg = np.transpose(preprocess_input(np.array(jpg, np.float64)), [2,0,1])
png = np.array(png)
png[png >= self.num_classes] = self.num_classes
# 转化成one_hot的形式
# 在这里需要+1是因为voc数据集有些标签具有白边部分
seg_labels = np.eye(self.num_classes + 1)[png.reshape([-1])]
seg_labels = seg_labels.reshape((int(self.input_shape[0]), int(self.input_shape[1]), self.num_classes + 1))
return jpg, png, seg_labels
②get_random_data变形时需要对两张图做同样的变换
if not random:
iw, ih = image.size
scale = min(w/iw, h/ih)
nw = int(iw*scale)
nh = int(ih*scale)
image = image.resize((nw,nh), Image.BICUBIC)
new_image = Image.new('RGB', [w, h], (128,128,128))
new_image.paste(image, ((w-nw)//2, (h-nh)//2))
label = label.resize((nw,nh), Image.NEAREST)
new_label = Image.new('L', [w, h], (0))
new_label.paste(label, ((w-nw)//2, (h-nh)//2))
return new_image, new_label
③collate_fn需要进行修改
def deeplab_dataset_collate(batch):
images = []
pngs = []
seg_labels = []
for img, png, labels in batch:
images.append(img)
pngs.append(png)
seg_labels.append(labels)
images = torch.from_numpy(np.array(images)).type(torch.FloatTensor)
pngs = torch.from_numpy(np.array(pngs)).long()
seg_labels = torch.from_numpy(np.array(seg_labels)).type(torch.FloatTensor)
return images, pngs, seg_labels
with open(train_annotation_path, encoding='utf-8') as f:
train_lines = f.readlines()
with open(val_annotation_path, encoding='utf-8') as f:
val_lines = f.readlines()
#获取数据集长度
num_train = len(train_lines)
num_val = len(val_lines)
这里一般检查数据集是否足够大,也可不检查
train_dataset = MyDataset(train_lines, input_shape, anchors, batch_size, num_classes, train = True)
val_dataset = MyDataset(val_lines, input_shape, anchors, batch_size, num_classes, train = False)
关于dataloader:一般有以下5个参数:
1.dataset:数据集对象,dataset型
2.batch_size:批大小,int型
3.shuffe:每一轮epoch是否重新洗牌,bool型
4.num_workers:多进程读取
5.drop_last:当样本不能被batch_size取整时,是否丢弃最后一批数据,bool型
gen = DataLoader(train_dataset, shuffle = shuffle, batch_size = batch_size, num_workers = num_workers, pin_memory=True,
drop_last=True, collate_fn=ssd_dataset_collate, sampler=train_sampler)
gen_val = DataLoader(val_dataset , shuffle = shuffle, batch_size = batch_size, num_workers = num_workers, pin_memory=True,
drop_last=True, collate_fn=ssd_dataset_collate, sampler=val_sampler)