yoloV4 训练数据整理
数据集类
Dataset
torch.utils.data.Dataset是代表数据集的抽象类。 您的自定义数据集应继承Dataset并覆盖以下方法:
- len,以便len(dataset)返回数据集的大小。
- __getitem__支持索引,以便可以使用dataset[i]获取第个样本
我们将在__init__中读取 csv,但将图像读取留给__getitem__。 由于所有图像不会立即存储在内存中,而是根据需要读取,因此可以提高存储效率。
DataLoader
torch.utils.data.DataLoader是提供所有这些功能的迭代器。 下面使用的参数应该清楚。 感兴趣的一个参数是collate_fn。 您可以使用collate_fn指定需要如何精确地分批样品。 但是,默认精度在大多数情况下都可以正常工作。
数据集code源码解读
annotation_path = '2007_train.txt'
#----------------------------------------------------------------------#
# 验证集的划分在train.py代码里面进行
# 2007_test.txt和2007_val.txt里面没有内容是正常的。训练不会使用到。
# 当前划分方式下,验证集和训练集的比例为1:9
#----------------------------------------------------------------------#
val_split = 0.1
with open(annotation_path) as f:
lines = f.readlines()
np.random.seed(10101)
np.random.shuffle(lines)
np.random.seed(None)
num_val = int(len(lines)*val_split)
num_train = len(lines) - num_val
train_dataset = YoloDataset(lines[:num_train], (input_shape[0], input_shape[1]), mosaic=mosaic, is_train=True)
val_dataset = YoloDataset(lines[num_train:], (input_shape[0], input_shape[1]), mosaic=False, is_train=False)
gen = DataLoader(train_dataset, shuffle=True, batch_size=Batch_size, num_workers=4, pin_memory=True,
drop_last=True, collate_fn=yolo_dataset_collate)
gen_val = DataLoader(val_dataset, shuffle=True, batch_size=Batch_size, num_workers=4,pin_memory=True,
drop_last=True, collate_fn=yolo_dataset_collate)
通过调用YoloDataset类实例,YoloDataset.getitem()方法
class YoloDataset(Dataset):
def __init__(self, train_lines, image_size, mosaic=True, is_train=True):
super(YoloDataset, self).__init__()
self.train_lines = train_lines
self.train_batches = len(train_lines)
self.image_size = image_size
self.mosaic = mosaic
self.flag = True
self.is_train = is_train
def __len__(self):
return self.train_batches
def rand(self, a=0, b=1):
return np.random.rand() * (b - a) + a
def get_random_data(self, annotation_line, input_shape, jitter=.3, hue=.1, sat=1.5, val=1.5, random=True):
"""实时数据增强的随机预处理"""
#['/home/quenin/soft/project/development/pytorch_yoloV4/VOCdevkit/VOC2007/JPEGImages/armed_police_coat_0611_v1_original_89_1.jpg','49,48,113,124,0', '115,57,176,129,0', '176,59,239,131,0']
line = annotation_line.split()
#读取图片
image = Image.open(line[0])
#图片原始大小 (209,313)
iw, ih = image.size
#resize (416,416)
h, w = input_shape
#(len(gt_boxes),5) 图片中打标的box 数量,(x0,yo,x1,y1,class_num) (x0,yo)为box左上角坐标,(x1,y1)为box右下角坐标;
#class_num 为类别编号
box = np.array([np.array(list(map(int, box.split(',')))) for box in line[1:]])
# letterbox_image 变换 图片 resize (416,416),图片保持原来的长宽比,不足的地方用(128,128,128)填充
#例子:(209,313) -->(277,416)保持长宽比--->(69+277+69,416) 填充(128,128,128)
#
if not random:
scale = min(w/iw, h/ih)
nw = int(iw*scale)
nh = int(ih*scale)
dx = (w-nw)//2
dy = (h-nh)//2
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)
# 调整目标框坐标
#新的目标框坐标和resize前目标框坐标框关系
# box[:, [0, 2]] = box[:, [0, 2]] * nw / iw + dx
# box[:, [1, 3]] = box[:, [1, 3]] * nh / ih + dy
box_data = np.zeros((len(box), 5))
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
# <0 数据重置为0
box[:, 0:2][box[:, 0:2] < 0] = 0
# 超过 416 的重置为416
box[:, 2][box[:, 2] > w] = w
box[:, 3][box[:, 3] > h] = h
box_w = box[:, 2] - box[:, 0]
box_h = box[:, 3] - box[:, 1]
# 保留有效框 长宽均大于1 的box
box = box[np.logical_and(box_w > 1, box_h > 1)] # 保留有效框
box_data = np.zeros((len(box), 5))
box_data[:len(box)] = box
return image_data, box_data
# 调整图片大小
new_ar = w / h * 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)
# 放置图片
dx = int(self.rand(0, w - nw))
dy = int(self.rand(0, h - nh))
new_image = Image.new('RGB', (w, h),
(np.random.randint(0, 255), np.random.randint(0, 255), np.random.randint(0, 255)))
new_image.paste(image, (dx, dy))
image = new_image
# 是否翻转图片
flip = self.rand() < .5
if flip:
image = image.transpose(Image.FLIP_LEFT_RIGHT)
# 色域变换
hue = self.rand(-hue, hue)
sat = self.rand(1, sat) if self.rand() < .5 else 1 / self.rand(1, sat)
val = self.rand(1, val) if self.rand() < .5 else 1 / self.rand(1, val)
x = cv2.cvtColor(np.array(image,np.float32)/255, cv2.COLOR_RGB2HSV)
x[..., 0] += hue*360
x[..., 0][x[..., 0]>1] -= 1
x[..., 0][x[..., 0]<0] += 1
x[..., 1] *= sat
x[..., 2] *= val
x[x[:,:, 0]>360, 0] = 360
x[:, :, 1:][x[:, :, 1:]>1] = 1
x[x<0] = 0
image_data = cv2.cvtColor(x, cv2.COLOR_HSV2RGB)*255
# 调整目标框坐标
box_data = np.zeros((len(box), 5))
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)] # 保留有效框
box_data = np.zeros((len(box), 5))
box_data[:len(box)] = box
return image_data, box_data
def get_random_data_with_Mosaic(self, annotation_line, input_shape, hue=.1, sat=1.5, val=1.5):
h, w = input_shape
min_offset_x = 0.3
min_offset_y = 0.3
scale_low = 1 - min(min_offset_x, min_offset_y)
scale_high = scale_low + 0.2
image_datas = []
box_datas = []
index = 0
place_x = [0, 0, int(w * min_offset_x), int(w * min_offset_x)]
place_y = [0, int(h * min_offset_y), int(h * min_offset_y), 0]
for line in annotation_line:
# 每一行进行分割
line_content = line.split()
# 打开图片
image = Image.open(line_content[0])
image = image.convert("RGB")
# 图片的大小
iw, ih = image.size
# 保存框的位置
box = np.array([np.array(list(map(int, box.split(',')))) for box in line_content[1:]])
# 是否翻转图片
flip = self.rand() < .5
if flip and len(box) > 0:
image = image.transpose(Image.FLIP_LEFT_RIGHT)
box[:, [0, 2]] = iw - box[:, [2, 0]]
# 对输入进来的图片进行缩放
new_ar = w / h
scale = self.rand(scale_low, scale_high)
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)
# 进行色域变换
hue = self.rand(-hue, hue)
sat = self.rand(1, sat) if self.rand() < .5 else 1 / self.rand(1, sat)
val = self.rand(1, val) if self.rand() < .5 else 1 / self.rand(1, val)
x = cv2.cvtColor(np.array(image,np.float32)/255, cv2.COLOR_RGB2HSV)
x[..., 0] += hue*360
x[..., 0][x[..., 0]>1] -= 1
x[..., 0][x[..., 0]<0] += 1
x[..., 1] *= sat
x[..., 2] *= val
x[x[:,:, 0]>360, 0] = 360
x[:, :, 1:][x[:, :, 1:]>1] = 1
x[x<0] = 0
image = cv2.cvtColor(x, cv2.COLOR_HSV2RGB) # numpy array, 0 to 1
image = Image.fromarray((image * 255).astype(np.uint8))
# 将图片进行放置,分别对应四张分割图片的位置
dx = place_x[index]
dy = place_y[index]
new_image = Image.new('RGB', (w, h),
(np.random.randint(0, 255), np.random.randint(0, 255), np.random.randint(0, 255)))
new_image.paste(image, (dx, dy))
image_data = np.array(new_image)
index = index + 1
box_data = []
# 对box进行重新处理
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)]
box_data = np.zeros((len(box), 5))
box_data[:len(box)] = box
image_datas.append(image_data)
box_datas.append(box_data)
# 将图片分割,放在一起
cutx = np.random.randint(int(w * min_offset_x), int(w * (1 - min_offset_x)))
cuty = np.random.randint(int(h * min_offset_y), int(h * (1 - min_offset_y)))
new_image = np.zeros([h, w, 3])
new_image[:cuty, :cutx, :] = image_datas[0][:cuty, :cutx, :]
new_image[cuty:, :cutx, :] = image_datas[1][cuty:, :cutx, :]
new_image[cuty:, cutx:, :] = image_datas[2][cuty:, cutx:, :]
new_image[:cuty, cutx:, :] = image_datas[3][:cuty, cutx:, :]
# 对框进行进一步的处理
new_boxes = np.array(merge_bboxes(box_datas, cutx, cuty))
return new_image, new_boxes
def __getitem__(self, index):
lines = self.train_lines
n = self.train_batches
index = index % n
if self.mosaic:
if self.flag and (index + 4) < n:
img, y = self.get_random_data_with_Mosaic(lines[index:index + 4], self.image_size[0:2])
else:
img, y = self.get_random_data(lines[index], self.image_size[0:2], random=self.is_train)
self.flag = bool(1-self.flag)
else:
img, y = self.get_random_data(lines[index], self.image_size[0:2], random=self.is_train)
if len(y) != 0:
# 从坐标转换成0~1的百分比
#(x0,y0,x1,x2) 左上角,右下角坐标
boxes = np.array(y[:, :4], dtype=np.float32)
#坐标值数据归一化 (x0/416,yo/416,x1/416,y1/416)
boxes[:, 0] = boxes[:, 0] / self.image_size[1]
boxes[:, 1] = boxes[:, 1] / self.image_size[0]
boxes[:, 2] = boxes[:, 2] / self.image_size[1]
boxes[:, 3] = boxes[:, 3] / self.image_size[0]
boxes = np.maximum(np.minimum(boxes, 1), 0)
#坐标数据值转换 (x0,yo,x1,y1) -> (xc,yc,w,h) 转换成box中心点坐标,和长宽
boxes[:, 2] = boxes[:, 2] - boxes[:, 0]
boxes[:, 3] = boxes[:, 3] - boxes[:, 1]
boxes[:, 0] = boxes[:, 0] + boxes[:, 2] / 2
boxes[:, 1] = boxes[:, 1] + boxes[:, 3] / 2
# 拼接坐标 + 类别
y = np.concatenate([boxes, y[:, -1:]], axis=-1)
img = np.array(img, dtype=np.float32)
tmp_inp = np.transpose(img / 255.0, (2, 0, 1))
tmp_targets = np.array(y, dtype=np.float32)
return tmp_inp, tmp_targets
函数的输出值
tmp_inp (batchsize,3,416,416)
tmp_targets (batchsize,len(gt_boxes),5) (xt,yt,w,h,class)