def __next__(self):
self.count += 1
# print('11')
if self.count == self.nB:
raise StopIteration
ia = self.count * self.batch_size #
ib = min((self.count + 1) * self.batch_size, self.nF)
# ia = self.count * 4
# ib = min((self.count + 1) * 4, self.nF2)
if self.multi_scale:
# Multi-Scale YOLO Training
height = random.choice(range(10, 20)) * 32 # 320 - 608 pixels
else:
# Fixed-Scale YOLO Training
height = self.height
img_all = []
labels_all = []
for index, files_index in enumerate(range(ia, ib)):
# if index >= 4:
# img_path = self.temp[self.shuffled_vector_2[files_index]]
# label_path = self.temp_label[self.shuffled_vector_2[files_index]]
# else:
img_path = self.img_files[self.shuffled_vector[files_index]]
# print(img_path)
# img_path = 'D:/00-Data/rtpose_datasets/COCO/images' + img_path
# print(img_path)
label_path = self.label_files[self.shuffled_vector[files_index]]
# print(label_path.split('/')[-1])
label_path = 'D:/01-ComputerVisionEntries/10-HumanDetection/PyTorch-YOLOv3-master/labels/train/' + label_path.split(
'/')[-1]
# print(label_path)
# face_path =
# print(label_path)
# print(label_path)
img = cv2.imread(img_path) # BGR
# print(img_path)
if img is None:
print(img_path)
continue
augment_hsv = True
if self.augment and augment_hsv:
# SV augmentation by 50%
fraction = 0.50
img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
S = img_hsv[:, :, 1].astype(np.float32)
V = img_hsv[:, :, 2].astype(np.float32)
a = (random.random() * 2 - 1) * fraction + 1
S *= a
if a > 1:
np.clip(S, a_min=0, a_max=255, out=S)
a = (random.random() * 2 - 1) * fraction + 1
V *= a
if a > 1:
np.clip(V, a_min=0, a_max=255, out=V)
img_hsv[:, :, 1] = S.astype(np.uint8)
img_hsv[:, :, 2] = V.astype(np.uint8)
cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img)
h, w, _ = img.shape
# print(h, w,_)
img, ratio, padw, padh = resize_square(img,
height=height,
color=(127.5, 127.5, 127.5))
# print(padw, padh)
# Load labels
if os.path.isfile(label_path):
labels0 = np.loadtxt(label_path,
dtype=np.float32).reshape(-1, 5)
# Normalized xywh to pixel xyxy format
labels0[:, 1] = (labels0[:, 1] + labels0[:, 3] / 2) # center x
labels0[:, 2] = (labels0[:, 2] + labels0[:, 4] / 2) # center y
labels = labels0.copy()
# x1,y1, x2,y2 #ratio = old/new
labels[:, 1] = ratio * w * (
labels0[:, 1] -
labels0[:, 3] / 2) + padw # 将归一化的坐标值--恢复--转换成resize之后的坐标值
labels[:, 2] = ratio * h * (labels0[:, 2] -
labels0[:, 4] / 2) + padh
labels[:, 3] = ratio * w * (labels0[:, 1] +
labels0[:, 3] / 2) + padw
labels[:, 4] = ratio * h * (labels0[:, 2] +
labels0[:, 4] / 2) + padh
else:
labels = np.array([])
# Augment image and labels
if self.augment:
img, labels, M = random_affine(img,
labels,
degrees=(-5, 5),
translate=(0.10, 0.10),
scale=(0.90, 1.10))
# print(img.shape)
plotFlag = False
if plotFlag:
import matplotlib.pyplot as plt
plt.figure(figsize=(10, 10)) if index == 0 else None
plt.subplot(4, 4, index + 1).imshow(img[:, :, ::-1])
plt.plot(labels[:, [1, 3, 3, 1, 1]].T,
labels[:, [2, 2, 4, 4, 2]].T, '.-')
plt.axis('off')
nL = len(labels)
if nL > 0:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5].copy()) / height
# print(labels[:, 1])
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip & (random.random() > 0.5):
img = np.fliplr(img)
if nL > 0:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip & (random.random() > 0.5):
img = np.flipud(img)
if nL > 0:
labels[:, 2] = 1 - labels[:, 2]
img_all.append(img)
labels_all.append(torch.from_numpy(labels))
# Normalize
img_all = np.stack(img_all)[:, :, :, ::-1].transpose(
0, 3, 1, 2) # Channel first and cv2 to pytorch
img_all = np.ascontiguousarray(img_all, dtype=np.float32)
# img_all -= self.rgb_mean
# img_all /= self.rgb_std
# print(img_all.shape)
img_all /= 255.0
return torch.from_numpy(img_all), labels_all
def __getitem__(self, index):
if self.image_weights:
index = self.indices[index]
img_path = self.img_files[index]
label_path = self.label_files[index]
hyp = self.hyp
# Load image
img = self.imgs[index]
if img is None:
img = cv2.imread(img_path) # BGR
assert img is not None, 'Image Not Found ' + img_path
r = self.img_size / max(img.shape) # size ratio
if self.augment and r < 1: # if training (NOT testing), downsize to inference shape
h, w, _ = img.shape
img = cv2.resize(img, (int(w * r), int(h * r)), interpolation=cv2.INTER_LINEAR) # INTER_LINEAR fastest
# Augment colorspace
augment_hsv = True
if self.augment and augment_hsv:
# SV augmentation by 50%
img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV) # hue, sat, val
S = img_hsv[:, :, 1].astype(np.float32) # saturation
V = img_hsv[:, :, 2].astype(np.float32) # value
a = random.uniform(-1, 1) * hyp['hsv_s'] + 1
b = random.uniform(-1, 1) * hyp['hsv_v'] + 1
S *= a
V *= b
img_hsv[:, :, 1] = S if a < 1 else S.clip(None, 255)
img_hsv[:, :, 2] = V if b < 1 else V.clip(None, 255)
cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img)
# Letterbox
h, w, _ = img.shape
if self.rect:
shape = self.batch_shapes[self.batch[index]]
img, ratiow, ratioh, padw, padh = letterbox(img, new_shape=shape, mode='rect')
else:
shape = self.img_size
img, ratiow, ratioh, padw, padh = letterbox(img, new_shape=shape, mode='square')
# Load labels
labels = []
if os.path.isfile(label_path):
x = self.labels[index]
if x is None: # labels not preloaded
with open(label_path, 'r') as f:
x = np.array([x.split() for x in f.read().splitlines()], dtype=np.float32)
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:, 1] = ratiow * w * (x[:, 1] - x[:, 3] / 2) + padw
labels[:, 2] = ratioh * h * (x[:, 2] - x[:, 4] / 2) + padh
labels[:, 3] = ratiow * w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:, 4] = ratioh * h * (x[:, 2] + x[:, 4] / 2) + padh
# Augment image and labels
if self.augment:
img, labels = random_affine(img, labels,
degrees=hyp['degrees'],
translate=hyp['translate'],
scale=hyp['scale'],
shear=hyp['shear'])
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() > 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() > 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Normalize
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img, dtype=np.float32) # uint8 to float32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
return torch.from_numpy(img), labels_out, img_path, (h, w)
def __next__(self):
self.count += 1
if self.count == self.nB:
raise StopIteration
ia = self.count * self.batch_size
ib = min((self.count + 1) * self.batch_size, self.nF)
if self.multi_scale:
# Multi-Scale YOLO Training
height = random.choice(range(10, 20)) * 32 # 320 - 608 pixels
else:
# Fixed-Scale YOLO Training
height = self.height
img_all = []
labels_all = []
for index, files_index in enumerate(range(ia, ib)):
img_path = self.img_files[self.shuffled_vector[files_index]]
label_path = self.label_files[self.shuffled_vector[files_index]]
img = cv2.imread(os.path.join("image_train", img_path)) # BGR
#print(img.shape)
if img is None:
continue
augment_hsv = True
if self.augment and augment_hsv:
# SV augmentation by 50%
fraction = 0.50
img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
S = img_hsv[:, :, 1].astype(np.float32)
V = img_hsv[:, :, 2].astype(np.float32)
a = (random.random() * 2 - 1) * fraction + 1
S *= a
if a > 1:
np.clip(S, a_min=0, a_max=255, out=S)
a = (random.random() * 2 - 1) * fraction + 1
V *= a
if a > 1:
np.clip(V, a_min=0, a_max=255, out=V)
img_hsv[:, :, 1] = S.astype(np.uint8)
img_hsv[:, :, 2] = V.astype(np.uint8)
cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img)
h, w, _ = img.shape
img, ratio, padw, padh = letterbox(img, height=height)
# Load labels
if os.path.isfile(os.path.join("labels_txt", label_path)):
with warnings.catch_warnings():
warnings.simplefilter("ignore")
labels0 = np.loadtxt(os.path.join("labels_txt",
label_path),
dtype=np.float32).reshape(-1, 5)
#print(os.path.join("data_train",label_path))
#data = np.loadtxt(myfile, unpack=True)
#labels0 = np.loadtxt(os.path.join("data_train",label_path), dtype=np.float32).reshape(-1, 5)
#print('='*50, os.path.join("labels_txt",label_path))
#print()
# Normalized xywh to pixel xyxy format
labels = labels0.copy()
#labels[:, 0] = np.int32(labels[:, 0])
labels[:, 1] = ratio * w * (labels0[:, 1] -
labels0[:, 3] / 2) + padw
labels[:, 2] = ratio * h * (labels0[:, 2] -
labels0[:, 4] / 2) + padh
labels[:, 3] = ratio * w * (labels0[:, 1] +
labels0[:, 3] / 2) + padw
labels[:, 4] = ratio * h * (labels0[:, 2] +
labels0[:, 4] / 2) + padh
#print(labels)
else:
print(os.path.join("data_train", label_path))
print("st wrong")
labels = np.array([])
# Augment image and labels
if self.augment:
img, labels, M = random_affine(img,
labels,
degrees=(-5, 5),
translate=(0.10, 0.10),
scale=(0.90, 1.10))
plotFlag = False
if plotFlag:
import matplotlib.pyplot as plt
plt.figure(figsize=(10, 10)) if index == 0 else None
plt.subplot(4, 4, index + 1).imshow(img[:, :, ::-1])
plt.plot(labels[:, [1, 3, 3, 1, 1]].T,
labels[:, [2, 2, 4, 4, 2]].T, '.-')
plt.axis('off')
nL = len(labels)
if nL > 0:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5].copy()) / height
# labels[:, 1:5] = xyxy2xywh(labels[:, 1:5].copy()) / height
# print(os.path.join("data_train",label_path))
# print(labels)
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip & (random.random() > 0.5):
img = np.fliplr(img)
if nL > 0:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip & (random.random() > 0.5):
img = np.flipud(img)
if nL > 0:
labels[:, 2] = 1 - labels[:, 2]
img_all.append(img)
labels_all.append(torch.from_numpy(labels))
# Normalize
img_all = np.stack(img_all)[:, :, :, ::-1].transpose(
0, 3, 1, 2) # BGR to RGB and cv2 to pytorch
img_all = np.ascontiguousarray(img_all, dtype=np.float32)
img_all /= 255.0
return torch.from_numpy(img_all), labels_all
def __next__(self):
self.count += 1
if self.count == self.nB:
raise StopIteration
ia = self.count * self.batch_size
ib = min((self.count + 1) * self.batch_size, self.nF)
img_all, labels_all, img_paths, img_shapes = [], [], [], []
for index, files_index in enumerate(range(ia, ib)):
img_path = self.img_files[self.shuffled_vector[files_index]]
label_path = self.label_files[self.shuffled_vector[files_index]]
img = cv2.imread(img_path) # BGR
assert img is not None, 'File Not Found ' + img_path
augment_hsv = True
if self.augment and augment_hsv:
# SV augmentation by 50%
fraction = 0.50
img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
S = img_hsv[:, :, 1].astype(np.float32)
V = img_hsv[:, :, 2].astype(np.float32)
a = (random.random() * 2 - 1) * fraction + 1
S *= a
if a > 1:
np.clip(S, a_min=0, a_max=255, out=S)
a = (random.random() * 2 - 1) * fraction + 1
V *= a
if a > 1:
np.clip(V, a_min=0, a_max=255, out=V)
img_hsv[:, :, 1] = S.astype(np.uint8)
img_hsv[:, :, 2] = V.astype(np.uint8)
cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img)
h, w, _ = img.shape
img, ratio, padw, padh = letterbox(img, height=self.img_size)
# Load labels
if os.path.isfile(label_path):
labels0 = np.loadtxt(label_path,
dtype=np.float32).reshape(-1, 5)
# Normalized xywh to pixel xyxy format
labels = labels0.copy()
labels[:, 1] = ratio * w * (labels0[:, 1] -
labels0[:, 3] / 2) + padw
labels[:, 2] = ratio * h * (labels0[:, 2] -
labels0[:, 4] / 2) + padh
labels[:, 3] = ratio * w * (labels0[:, 1] +
labels0[:, 3] / 2) + padw
labels[:, 4] = ratio * h * (labels0[:, 2] +
labels0[:, 4] / 2) + padh
else:
labels = np.array([])
# Augment image and labels
if self.augment:
img, labels, M = random_affine(img,
labels,
degrees=(-5, 5),
translate=(0.10, 0.10),
scale=(0.90, 1.10))
plotFlag = False
if plotFlag:
import matplotlib.pyplot as plt
plt.figure(figsize=(10, 10)) if index == 0 else None
plt.subplot(4, 4, index + 1).imshow(img[:, :, ::-1])
plt.plot(labels[:, [1, 3, 3, 1, 1]].T,
labels[:, [2, 2, 4, 4, 2]].T, '.-')
plt.axis('off')
nL = len(labels)
if nL > 0:
# convert xyxy to xywh
labels[:,
1:5] = xyxy2xywh(labels[:, 1:5].copy()) / self.img_size
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip & (random.random() > 0.5):
img = np.fliplr(img)
if nL > 0:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip & (random.random() > 0.5):
img = np.flipud(img)
if nL > 0:
labels[:, 2] = 1 - labels[:, 2]
if nL > 0:
labels = np.concatenate((np.zeros(
(nL, 1), dtype='float32') + index, labels), 1)
labels_all.append(labels)
img_all.append(img)
img_paths.append(img_path)
img_shapes.append((h, w))
# Normalize
img_all = np.stack(img_all)[:, :, :, ::-1].transpose(
0, 3, 1, 2) # BGR to RGB and cv2 to pytorch
img_all = np.ascontiguousarray(img_all, dtype=np.float32)
img_all /= 255.0
labels_all = torch.from_numpy(np.concatenate(labels_all, 0))
return torch.from_numpy(img_all), labels_all, img_paths, img_shapes
def __getitem__(self, index):
img_path = self.img_files[index]
label_path = self.label_files[index]
img = cv2.imread(img_path) # BGR
assert img is not None, 'File Not Found ' + img_path
augment_hsv = True
if self.augment and augment_hsv:
# SV augmentation by 50%
fraction = 0.50 # must be < 1.0
img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
S = img_hsv[:, :, 1].astype(np.float32)
V = img_hsv[:, :, 2].astype(np.float32)
a = (random.random() * 2 - 1) * fraction + 1
S *= a
if a > 1:
np.clip(S, None, 255, out=S)
a = (random.random() * 2 - 1) * fraction + 1
V *= a
if a > 1:
np.clip(V, None, 255, out=V)
img_hsv[:, :, 1] = S # .astype(np.uint8)
img_hsv[:, :, 2] = V # .astype(np.uint8)
cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img)
h, w, _ = img.shape
img, ratio, padw, padh = letterbox(img, height=self.img_size)
# Load labels
labels = []
if os.path.isfile(label_path):
with open(label_path, 'r') as file:
lines = file.read().splitlines()
x = np.array([x.split() for x in lines], dtype=np.float32)
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:, 1] = ratio * w * (x[:, 1] - x[:, 3] / 2) + padw
labels[:, 2] = ratio * h * (x[:, 2] - x[:, 4] / 2) + padh
labels[:, 3] = ratio * w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:, 4] = ratio * h * (x[:, 2] + x[:, 4] / 2) + padh
# Augment image and labels
if self.augment:
img, labels = random_affine(img,
labels,
degrees=(-10, 10),
translate=(0.10, 0.10),
scale=(0.80, 1.20))
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5]) / self.img_size
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() > 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() > 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Normalize
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img, dtype=np.float32) # uint8 to float32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
return torch.from_numpy(img), labels_out, img_path, (h, w)
def __getitem__(self, index):
# 新的下角标
if self.image_weights:
index = self.indices[index]
img_path = self.img_files[index]
label_path = self.label_files[index]
hyp = self.hyp
mosaic = False and self.augment
# 如果开启镶嵌增强、数据增强
# 加载四张图片，作为一个镶嵌，具体看下文解析。
if mosaic:
# 加载镶嵌内容
img, labels = load_mosaic(self, index)
shapes = None
else:
# 加载图片
img, (h0, w0), (h, w) = load_image(self, index)
# 仿射变换
shape = self.batch_shapes[self.batch[
index]] if self.rect else self.img_size
img, ratio, pad = letterbox(img,
shape,
auto=False,
scaleup=self.augment)
shapes = (h0, w0), (
(h / h0, w / w0), pad)
# 加载标注文件
labels = []
if os.path.isfile(label_path):
x = self.labels[index]
if x is None: # 如果标签没有加载，读取label_path内容
with open(label_path, 'r') as f:
x = np.array(
[x.split() for x in f.read().splitlines()],
dtype=np.float32)
if x.size > 0:
# 将归一化后的xywh转化为左上角、右下角的表达形式
labels = x.copy()
labels[:, 1] = ratio[0] * w * (
x[:, 1] - x[:, 3] / 2) + pad[0] # pad width
labels[:, 2] = ratio[1] * h * (
x[:, 2] - x[:, 4] / 2) + pad[1] # pad height
labels[:, 3] = ratio[0] * w * (x[:, 1] +
x[:, 3] / 2) + pad[0]
labels[:, 4] = ratio[1] * h * (x[:, 2] +
x[:, 4] / 2) + pad[1]
if self.augment:
# 图片空间的数据增强
if not mosaic:
# 如果没有使用镶嵌的方法，那么对图片进行随机放射
img, labels = random_affine(img,
labels,
degrees=hyp['degrees'],
translate=hyp['translate'],
scale=hyp['scale'],
shear=hyp['shear'])
# 增强hsv空间
augment_hsv(img,
hgain=hyp['hsv_h'],
sgain=hyp['hsv_s'],
vgain=hyp['hsv_v'])
nL = len(labels) # 标注文件个数
if nL:
# 将 xyxy 格式转化为 xywh 格式
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])# x1, y1, x2, y2
# 归一化到0-1之间
labels[:, [2, 4]] /= img.shape[0] # height y1/h , y2/h
labels[:, [1, 3]] /= img.shape[1] # width x1/h , x2/h
if self.augment:
# 随机左右翻转
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# 随机上下翻转
ud_flip = False
if ud_flip and random.random() < 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# 图像维度转换
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img)
return torch.from_numpy(img), labels_out, img_path, shapes
def __getitem__(self, index):
img = self.load_image(index)
anns, K = self.load_annotations(index)
# Change h, w -> w, h
size = np.array([i for i in img.shape[:-1]], np.float32)[::-1]
center = np.array([i/2 for i in img.shape[:-1]], np.float32)[::-1]
# return img, K
"""
resize, horizontal flip, and affine augmentation are performed here.
since it is complicated to compute heatmap w.r.t transform.
"""
flipped = False
if (self.is_train) and (np.random.rand() < self.flip_prob):
flipped = True
img = cv2.flip(img, 1)
center[0] = size[0] - center[0] - 1
K[0, 2] = size[0] - K[0, 2] - 1
affine = False
if (self.is_train) and (np.random.rand() < self.aug_prob):
img, target, trans_mat = random_affine(img, degrees=0, translate=.1, scale=.1)
affine = True
'''
TODO: affine the label mat
point = affine_transform(point, trans_mat)
box2d[:2] = affine_transform(box2d[:2], trans_mat)
box2d[2:] = affine_transform(box2d[2:], trans_mat)
TODO:There is something wrong when clip after resize
box2d[[0, 2]] = box2d[[0, 2]].clip(0, self.input_width - 1)
box2d[[1, 3]] = box2d[[1, 3]].clip(0, self.input_height - 1)
'''
resize = False
if img.shape[0] != self.input_height | img.shape[1] != self.input_width:
img, ratio, pad = resize_image_with_pad(img, (self.input_height, self.input_width))
resize = True
labels = np.zeros((len(anns), 9))
for i, a in enumerate(anns):
a = a.copy()
_cls = a["label"]
locs = np.array(a["locations"])
rot_y = np.array(a["rot_y"])
if flipped:
locs[0] *= -1
rot_y *= -1
# We can get 2D bbox by labels or calculate by camera&3D bbox directly
point, box2d, box3d = encode_label(
K, rot_y, a["dimensions"], locs
)
# 当图像中的物体不全时，计算出的2D框会超出图像大小范围，这里先使用标注信息替代
box2d = a["bbox"]
labels[i, 0] = _cls
labels[i, 1:5] = np.array(box2d)
labels[i, 5:8] = np.array(a["dimensions"])
labels[i, 8] = rot_y
# h, w = box2d[3] - box2d[1], box2d[2] - box2d[0]
nL = len(labels)
if nL > 0:
if resize:
labels[:, 1] = ratio[0] * labels[:, 1] + pad[0] # pad width
labels[:, 2] = ratio[1] * labels[:, 2] + pad[1] # pad height
labels[:, 3] = ratio[0] * labels[:, 3] + pad[0]
labels[:, 4] = ratio[1] * labels[:, 4] + pad[1]
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
#
labels[labels < 0] = 0.0
label_out = torch.zeros((nL, self.out_parms+1))
if nL > 0:
label_out[:, 1:] = torch.from_numpy(labels)
# Convert
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img)
shapes = (size[1], size[0]), (size[1]/img.shape[2], size[0]/img.shape[1], pad)
return torch.from_numpy(img), label_out[:, :6], self.files[index], shapes
def __next__(self):
self.count += 1
if self.count == self.nB:
raise StopIteration
ia = self.count * self.batch_size
ib = min((self.count + 1) * self.batch_size, self.nF)
height = self.height
# height = random.choice([15, 17, 19, 21]) * 32
img_all = []
labels_all = []
for index, files_index in enumerate(range(ia, ib)):
# img_path = self.files[self.shuffled_vector[files_index]] # BGR
img_path = '%s/%g.bmp' % (self.path,
self.shuffled_vector[files_index])
# img_path = '/Users/glennjocher/Downloads/DATA/xview/train_images/2294.bmp'
img0 = cv2.imread(img_path)
if img0 is None:
continue
augment_hsv = True
if augment_hsv:
# SV augmentation by 50%
fraction = 0.50
img_hsv = cv2.cvtColor(img0, cv2.COLOR_BGR2HSV)
S = img_hsv[:, :, 1].astype(np.float32)
V = img_hsv[:, :, 2].astype(np.float32)
a = (random.random() * 2 - 1) * fraction + 1
S *= a
if a > 1:
np.clip(S, a_min=0, a_max=255, out=S)
a = (random.random() * 2 - 1) * fraction + 1
V *= a
if a > 1:
np.clip(V, a_min=0, a_max=255, out=V)
img_hsv[:, :, 1] = S.astype(np.uint8)
img_hsv[:, :, 2] = V.astype(np.uint8)
cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img0)
# Load labels
chip = img_path.rsplit('/')[-1]
i = (self.mat['id'] == float(
chip.replace('.tif', '').replace('.bmp', ''))).nonzero()[0]
labels1 = self.mat['targets'][i]
# Remove buildings and small cars
# labels1 = labels1[(labels1[:, 0] != 5) & (labels1[:, 0] != 48)]
img1, labels1, M = random_affine(img0,
targets=labels1,
degrees=(-20, 20),
translate=(0.01, 0.01),
scale=(0.70, 1.30)) # RGB
nL1 = len(labels1)
border = height / 2 + 1
# Pick 100 random points inside image
r = np.ones((100, 3))
r[:, :2] = np.random.rand(
100, 2) * (np.array(img0.shape)[[1, 0]] - border * 2) + border
r = (r @ M.T)[:, :2]
r = r[np.all(r > border, 1)
& np.all(r < img1.shape[0] - border, 1)]
# import matplotlib.pyplot as plt
# plt.imshow(img1[:, :, ::-1])
# plt.plot(labels1[:, [1, 3, 3, 1, 1]].T, labels1[:, [2, 2, 4, 4, 2]].T, '.-')
# plt.plot(r[:,0],r[:,1],'.')
if nL1 > 0:
weights = []
for k in range(len(r)):
x = (labels1[:, 1] + labels1[:, 3]) / 2
y = (labels1[:, 2] + labels1[:, 4]) / 2
c = labels1[(abs(r[k, 0] - x) < height / 2) &
(abs(r[k, 1] - y) < height / 2), 0]
if len(c) == 0:
weights.append(1e-16)
else:
weights.append(self.class_weights[c.astype(
np.int8)].sum())
weights = np.array(weights)
weights /= weights.sum()
r = r[np.random.choice(len(r),
size=8,
p=weights,
replace=False)]
if nL1 > 0:
area0 = (labels1[:, 3] - labels1[:, 1]) * (labels1[:, 4] -
labels1[:, 2])
h, w, _ = img1.shape
for j in range(8):
labels = np.array([], dtype=np.float32)
pad_x = int(r[j, 0] - height / 2)
pad_y = int(r[j, 1] - height / 2)
if nL1 > 0:
labels = labels1.copy()
labels[:, [1, 3]] -= pad_x
labels[:, [2, 4]] -= pad_y
np.clip(labels[:, 1:5], 0, height, out=labels[:, 1:5])
lw = labels[:, 3] - labels[:, 1]
lh = labels[:, 4] - labels[:, 2]
area = lw * lh
ar = np.maximum(lw / (lh + 1e-16), lh / (lw + 1e-16))
# objects must have width and height > 4 pixels
labels = labels[(lw > 4) & (lh > 4) & (area > 20) &
(area / area0 > 0.1) & (ar < 10)]
# pad_x, pad_y, counter = 0, 0, 0
# while (counter < len(r)) & (len(labels) == 0):
# pad_x = int(r[counter, 0] - height / 2)
# pad_y = int(r[counter, 1] - height / 2)
#
# if nL1 == 0:
# break
#
# labels = labels1.copy()
# labels[:, [1, 3]] -= pad_x
# labels[:, [2, 4]] -= pad_y
# labels[:, 1:5] = np.clip(labels[:, 1:5], 0, height)
#
# lw = labels[:, 3] - labels[:, 1]
# lh = labels[:, 4] - labels[:, 2]
# area = lw * lh
# ar = np.maximum(lw / (lh + 1e-16), lh / (lw + 1e-16))
#
# # objects must have width and height > 4 pixels
# labels = labels[(lw > 4) & (lh > 4) & (area / area0 > 0.2) & (ar < 15)]
# counter += 1
img = img1[pad_y:pad_y + height, pad_x:pad_x + height]
# import matplotlib.pyplot as plt
# plt.subplot(4, 4, j+1).imshow(img[:, :, ::-1])
# plt.plot(labels[:, [1, 3, 3, 1, 1]].T, labels[:, [2, 2, 4, 4, 2]].T, '.-')
nL = len(labels)
if nL > 0:
# convert labels to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5].copy()) / height
# remap xview classes 11-94 to 0-61
# labels[:, 0] = xview_classes2indices(labels[:, 0])
# random lr flip
if random.random() > 0.5:
img = np.fliplr(img)
if nL > 0:
labels[:, 1] = 1 - labels[:, 1]
# random ud flip
if random.random() > 0.5:
img = np.flipud(img)
if nL > 0:
labels[:, 2] = 1 - labels[:, 2]
img_all.append(img)
labels_all.append(torch.from_numpy(labels))
# Randomize
i = np.random.permutation(len(labels_all))
img_all = [img_all[j] for j in i]
labels_all = [labels_all[j] for j in i]
# Normalize
img_all = np.stack(img_all)[:, :, :, ::-1].transpose(
0, 3, 1, 2) # BGR to RGB and cv2 to pytorch
img_all = np.ascontiguousarray(img_all, dtype=np.float32)
img_all -= self.rgb_mean
img_all /= self.rgb_std
return torch.from_numpy(img_all), labels_all
def detect(input_image, save_img=True):
out, source, weights, view_img, save_txt, imgsz = \
opt.output, opt.source, opt.weights, opt.view_img, opt.save_txt, opt.img_size
global frame_num, model
global pubCentBlem, pubCentUnBlem
# Initialize
#device = torch_utils.select_device(opt.device)
#if os.path.exists(out):
# shutil.rmtree(out) # delete output folder
#os.makedirs(out) # make new output folder
half = device.type != 'cpu' # half precision only supported on CUDA
if half:
model.half() # to FP16
# Get names and colors
names = model.module.names if hasattr(model, 'module') else model.names
print("class names array ", names)
# names = ['blemished', 'unblemished', 'glove', 'belt', 'bin', 'head']
colors = [[random.randint(0, 255) for _ in range(3)]
for _ in range(len(names))]
# Run inference
t0 = time.time()
img = torch.zeros((1, 3, imgsz, imgsz), device=device) # init img
_ = model(img.half() if half else img
) if device.type != 'cpu' else None # run once
bounding_boxes_all_images = []
img0 = input_image.astype('float32')
#img0 = cv2.resize(img0, (640,480), interpolation = cv2.INTER_AREA)
img = letterbox(img0, new_shape=imgsz)[0]
# Convert
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img)
dataset = [("frame_num" + str(frame_num) + '.jpg', img, img0, None)]
for path, img, im0s, vid_cap in dataset:
img = torch.from_numpy(img).to(device)
img = img.half() if half else img.float() # uint8 to fp16/32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
t1 = torch_utils.time_synchronized()
pred = model(img, augment=opt.augment)[0]
# pred = model(img)[0]
t2 = torch_utils.time_synchronized()
# Apply NMS
pred = non_max_suppression(pred,
opt.conf_thres,
opt.iou_thres,
classes=opt.classes,
agnostic=opt.agnostic_nms)
t3 = torch_utils.time_synchronized()
# needed specific to each image
Loc2arrCent = {}
arrCentBlem = []
arrCentUnBlem = []
Loc2Cls = {}
# Process detections
for i, det in enumerate(pred): # detections per image
p, s, im0 = path, '', im0s
save_path = str(Path(out) / Path(p).name)
s += '%gx%g ' % img.shape[2:] # print string
gn = torch.tensor(im0.shape)[[1, 0, 1,
0]] # normalization gain whwh
bounding_boxes = {}
if det is not None and len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4],
im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
s += '%g %ss, ' % (n, names[int(c)]) # add to string
# Write results
minx = 5000
miny = 5000
maxx = 0
maxy = 0
box_num = 0
bounding_boxes = {}
for *xyxy, conf, cls in det:
box_num += 1
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
#print("Box num:", box_num, " label: ",names[int(cls)], " xyxy: ", int(xyxy[0]), int(xyxy[1]), int(xyxy[2]), int(xyxy[3]), "xywh: ", float(xywh[0]), float(xywh[1]), float(xywh[2]), float(xywh[3]))
if (bounding_boxes.get(names[int(cls)], None) == None):
bounding_boxes[names[int(cls)]] = [[
int(xyxy[0]),
int(xyxy[1]),
abs(int(xyxy[2]) - int(xyxy[0])),
abs(int(xyxy[3]) - int(xyxy[1]))
]]
else:
bounding_boxes[names[int(cls)]].append([
int(xyxy[0]),
int(xyxy[1]),
abs(int(xyxy[2]) - int(xyxy[0])),
abs(int(xyxy[3]) - int(xyxy[1]))
])
tlx, tly, brx, bry = int(xyxy[0]), int(xyxy[1]), int(
xyxy[2]), int(xyxy[3])
if tlx < minx:
minx = tlx
if tly < miny:
miny = tly
if bry > maxy:
maxy = bry
if brx > maxx:
maxx = brx #crop_img = img[y:y+h, x:x+w]
if save_txt: # Write to file
xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) /
gn).view(-1).tolist() # normalized xywh
with open(save_path[:save_path.rfind('.')] + '.txt',
'a') as file:
file.write(('%g ' * 5 + '\n') %
(cls, *xywh)) # label format
if save_img or view_img: # Add bbox to image
label = '%s %.2f' % (names[int(cls)], conf)
# print("label input to plot_one_box ",label)
# plot_one_box(xyxy, im0, label=label, color=colors[int(cls)], line_thickness=3)
plot_one_box(xyxy,
im0,
label=label,
color=colors[int(cls)],
line_thickness=1)
# append onion centroid to list of centroids
if names[int(cls)] == 'blemished' or names[int(
cls)] == 'unblemished':
centx = (tlx + brx) / 2
centy = (tly + bry) / 2
if centx > 250:
# FOR CAMERA LOOKING FROM FRONT
# if center is on the belt
if names[int(cls)] == 'blemished':
Loc2Cls[centx] = 0
else:
Loc2Cls[centx] = 1
Loc2arrCent[centx] = (centx, centy)
if names[int(
cls
)] == 'blemished' and centx not in arrCentBlem:
arrCentBlem.append(centx)
arrCentBlem.append(centy)
else:
if centx not in arrCentUnBlem:
arrCentUnBlem.append(centx)
arrCentUnBlem.append(centy)
############################################### all bounding boxes for this frames: "bounding_boxes" ########################################################################################################
bounding_boxes_all_images.append(bounding_boxes)
print("Frame number = ", frame_num, "Bounding boxes: ",
bounding_boxes)
frame_num += 1
# Print time (inference + NMS)
print('%s Inference. (%.3fs)' % (s, t2 - t1))
print('%s NMS. (%.3fs)' % (s, t3 - t2))
#print('im0.shape before',im0.shape)
############################################### cropping
#im0 = im0[miny:maxy, minx:maxx]
#print('im0.shape after',im0.shape)
# Stream results
view_img = True
if view_img:
#cv2_imshow( im0)
not_showing = True
#cv2.imshow(p, im0)
if cv2.waitKey(1) == ord('q'): # q to quit
raise StopIteration
# Save results (image with detections)
save_img = True
if save_img: #'src/beginner_tutorials/scripts/yolov5/inference/output'
#save_path = '/home/psuresh/src/beginner_tutorials/scripts/yolov5/inference/output/frame1.jpg'
cv2.imwrite(save_path, im0) #if dataset.mode == 'images':
########################################################## comment this line to avoid saving images ########################################################################################################
# publish centroids for current image
print("publishing arrCentBlem ", arrCentBlem)
print("publishing arrCentUnBlem ", arrCentUnBlem)
pubCentBlem.publish(Float32MultiArray(data=arrCentBlem))
pubCentUnBlem.publish(Float32MultiArray(data=arrCentUnBlem))
# make array LocOrderedPreds
LocOrderedPreds = list(
OrderedDict(sorted(Loc2Cls.items(), key=lambda t: t[0])).values())
print("publishing LocOrderedPreds ", LocOrderedPreds)
pubLocOrderedPreds.publish(Int32MultiArray(data=LocOrderedPreds))
arrCentxyOrd = list(
OrderedDict(sorted(Loc2arrCent.items(),
key=lambda t: t[0])).values())
print("arrCentxyOrd ", arrCentxyOrd)
arrCentLocOrd = []
for i in range(len(arrCentxyOrd)):
arrCentLocOrd.append(arrCentxyOrd[i][0])
arrCentLocOrd.append(arrCentxyOrd[i][1])
print("publishing arrCentLocOrd ", arrCentLocOrd)
pubCentLocOrd.publish(Float32MultiArray(data=arrCentLocOrd))
#global img_processed
#img_processed = True
if save_txt or save_img:
print('Results saved to %s' % os.getcwd() + os.sep + out)
if platform == 'darwin': # MacOS
os.system('open ' + save_path)
print('Done. (%.3fs)' % (time.time() - t0))
return bounding_boxes_all_images
def detect(self, outfile=None):
frame_cnt = -1
if outfile is not None:
f = open(outfile, 'w')
print("begin....")
while self.vdo.grab():
frame_cnt += 1
if frame_cnt % 3 == 0:
continue
start = time.time()
_, ori_im = self.vdo.retrieve()
im = ori_im
t1_begin = time.time()
bbox_xxyy, cls_conf, cls_ids = self.yolo3.predict(im)
t1_end = time.time()
t2_begin = time.time()
if bbox_xxyy is not None:
# select class
# mask = cls_ids == 0
# bbox_xxyy = bbox_xxyy[mask]
# bbox_xxyy[:, 3:] *= 1.2
# cls_conf = cls_conf[mask]
bbox_xcycwh = xyxy2xywh(bbox_xxyy)
outputs = self.deepsort.update(bbox_xcycwh, cls_conf, im)
if len(outputs) > 0:
bbox_xyxy = outputs[:, :4]
identities = outputs[:, -1]
# 画框
ori_im = draw_bboxes(ori_im, bbox_xyxy, identities)
# frame, id, tlwh(%.2f),1,-1,-1,-1
if outfile is not None:
box_xywh = xyxy2tlwh(bbox_xyxy)
for i in range(len(box_xywh)):
write_line = "%d,%d,%d,%d,%d,%d,1,-1,-1,-1\n" % (
frame_cnt + 1, outputs[i, -1],
int(box_xywh[i][0]), int(box_xywh[i][1]),
int(box_xywh[i][2]), int(box_xywh[i][3]))
f.write(write_line)
t2_end = time.time()
end = time.time()
print(
"frame:%d|det:%.4f|sort:%.4f|total:%.4f|det p:%.2f%%|fps:%.2f"
% (frame_cnt, (t1_end - t1_begin), (t2_end - t2_begin),
(end - start), ((t1_end - t1_begin) * 100 /
((end - start))), (1 / (end - start))))
if self.args.display:
cv2.imshow("test", ori_im)
cv2.waitKey(1)
if self.args.save_path:
self.output.write(ori_im)
if outfile is not None:
f.close()
def __getitem__(self, index):
img_path = self.img_files[index]
label_path = self.label_files[index]
# Load image
img, (h0, w0), (h, w) = load_image(self, index)
# Letterbox
shape = self.batch_shapes[self.batch[
index]] if self.rect else self.img_size # final letterboxed shape
img, ratio, pad = letterbox(img,
shape,
auto=False,
scaleup=self.augment)
shapes = (h0, w0), ((h / h0, w / w0), pad) # for COCO mAP rescaling
# Load labels
labels = []
if os.path.isfile(label_path):
x = self.labels[index]
if x is None: # labels not preloaded
with open(label_path, 'r') as f:
x = np.array([x.split() for x in f.read().splitlines()],
dtype=np.float32)
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:,
1] = ratio[0] * w * (x[:, 1] -
x[:, 3] / 2) + pad[0] # pad width
labels[:,
2] = ratio[1] * h * (x[:, 2] -
x[:, 4] / 2) + pad[1] # pad height
labels[:, 3] = ratio[0] * w * (x[:, 1] + x[:, 3] / 2) + pad[0]
labels[:, 4] = ratio[1] * h * (x[:, 2] + x[:, 4] / 2) + pad[1]
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() < 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Convert
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img)
# print (f'before loader {torch.from_numpy(img).shape}_{labels_out}_{img_path}')
return torch.from_numpy(img), labels_out, img_path, shapes
def __getitem__(self, index):
img_uri = self.img_files[index]
img_labels = self.labels[index]
# don't download, since it was already downloaded in the init
img_path = img_uri
img_name = ("_".join(map(str, img_path.split("_")[-5:])))
orig_img = PIL.Image.open(img_path).convert('RGB')
if orig_img is None:
raise Exception(
"Empty image: {img_path}".format(img_path=img_path))
if self.vis_batch and len(img_labels) > 0:
vis_orig_img = copy.deepcopy(orig_img)
labels = add_class_dimension_to_labels(img_labels)
labels = xyhw2xyxy_corner(labels, skip_class_dimension=True)
tmp_path = os.path.join(visualization_tmp_path,
img_name[:-4] + ".jpg")
visualize_and_save_to_local(vis_orig_img,
labels,
tmp_path,
box_color="green")
print(f'new image uploaded to {tmp_path}')
# First, handle image re-shaping
if self.ts:
scale = self.scales[index]
scaled_img = scale_image(orig_img, scale)
scaled_img_width, scaled_img_height = scaled_img.size
patch_width, patch_height = self.width, self.height
vert_pad, horiz_pad = pre_tile_padding(scaled_img_width,
scaled_img_height,
patch_width, patch_height)
padded_img = torchvision.transforms.functional.pad(
scaled_img,
padding=(horiz_pad, vert_pad, horiz_pad, vert_pad),
fill=(127, 127, 127),
padding_mode="constant")
padded_img_width, padded_img_height = padded_img.size
_, _, n_patches, _, _ = get_patch_spacings(padded_img_width,
padded_img_height,
patch_width,
patch_height)
patch_index = random.randint(0, n_patches - 1)
if self.debug_mode:
patch_index = 0
img, boundary = get_patch(padded_img, patch_width, patch_height,
patch_index)
else:
orig_img_width, orig_img_height = orig_img.size
vert_pad, horiz_pad, ratio = calculate_padding(
orig_img_height, orig_img_width, self.height, self.width)
img = torchvision.transforms.functional.pad(
orig_img,
padding=(horiz_pad, vert_pad, horiz_pad, vert_pad),
fill=(127, 127, 127),
padding_mode="constant")
img = torchvision.transforms.functional.resize(
img, (self.height, self.width))
# If no labels, no need to do augmentation (this should change in the future)
# so immediately return with the padded image and empty labels
if len(img_labels) == 0:
labels = torch.zeros((len(img_labels), 5))
img = torchvision.transforms.functional.to_tensor(img)
labels = F.pad(
labels,
pad=[0, 0, 0, self.num_targets_per_image - len(labels)],
mode="constant")
return img_uri, img, labels
# Next, handle label re-shaping
labels = add_class_dimension_to_labels(img_labels)
labels = xyhw2xyxy_corner(labels)
if self.ts:
labels = scale_labels(labels, self.scales[index])
labels = add_padding_on_each_side(labels, horiz_pad, vert_pad)
if self.vis_batch:
tmp_path = os.path.join(visualization_tmp_path,
img_name[:-4] + "_scaled.jpg")
visualize_and_save_to_local(padded_img,
labels,
tmp_path,
box_color="red")
labels_temp = filter_and_offset_labels(labels, boundary)
if self.vis_batch:
pre_vis_labels = copy.deepcopy(labels)
for i in range(n_patches):
vis_patch_img, boundary = get_patch(
padded_img, patch_width, patch_height, i)
labels = filter_and_offset_labels(pre_vis_labels, boundary)
tmp_path = os.path.join(visualization_tmp_path, img_name[:-4] + \
"_patch_{}.jpg".format(i))
visualize_and_save_to_local(vis_patch_img,
labels,
tmp_path,
box_color="blue")
if self.upload_dataset:
pre_vis_labels = copy.deepcopy(labels)
for i in range(n_patches):
vis_patch_img, boundary = get_patch(
padded_img, patch_width, patch_height, i)
labels = filter_and_offset_labels(pre_vis_labels, boundary)
tmp_path = os.path.join(visualization_tmp_path, img_name[:-4] + \
"_patch_{}.jpg".format(i))
upload_label_and_image_to_gcloud(vis_patch_img, labels,
tmp_path)
else:
labels = filter_and_offset_labels(labels, boundary)
else:
labels = add_padding_on_each_side(labels, horiz_pad, vert_pad)
labels = scale_labels(labels, ratio)
labels_temp = labels
if self.vis_batch:
tmp_path = os.path.join(visualization_tmp_path,
img_name[:-4] + "_pad_resized.jpg")
visualize_and_save_to_local(img,
labels,
tmp_path,
box_color="blue")
labels = labels_temp
if self.vis_batch and self.data_aug:
vis_aug_img = copy.deepcopy(img)
tmp_path = os.path.join(visualization_tmp_path,
img_name[:-4] + "_before_aug.jpg")
visualize_and_save_to_local(vis_aug_img,
labels,
tmp_path,
box_color="red")
if self.augment_hsv or self.data_aug:
if random.random() > 0.5:
img = self.jitter(img)
# no transformation on labels
# Augment image and labels
img_width, img_height = img.size
if self.augment_affine or self.data_aug:
if random.random() > 0:
angle = random.uniform(-10, 10)
translate = (random.uniform(-40,
40), random.uniform(-40,
40)) ## WORKS
scale = random.uniform(0.9, 1.1)
shear = random.uniform(-3, 3)
img = torchvision.transforms.functional.affine(img,
angle,
translate,
scale,
shear,
2,
fillcolor=(127,
127,
127))
labels = affine_labels(img_height, img_width, labels, -angle,
translate, scale, (-shear, 0))
if self.bw:
img = torchvision.transforms.functional.to_grayscale(
img, num_output_channels=1)
# random left-right flip
if self.lr_flip:
if random.random() > 0.5:
img = torchvision.transforms.functional.hflip(img)
# Is this correct?
# Not immediately obvious, when composed with the angle shift above
labels[:, 1] = img_width - labels[:, 1]
labels[:, 3] = img_width - labels[:, 3]
# GaussianBlur, needs further development
if self.blur:
if random.random() > 0.2:
arr = np.asarray(img)
angle = random.uniform(40, -40)
sigma = random.uniform(0, 3.00)
seq = iaa.Sequential([iaa.GaussianBlur(sigma=sigma)])
images_aug = seq.augment_images(arr)
img = PIL.Image.fromarray(np.uint8(images_aug), 'RGB')
#AdditiveGaussianNoise
if self.noise:
if random.random() > 0.3:
arr = np.asarray(img)
scale = random.uniform(0, 0.03 * 255)
seq = iaa.Sequential([
iaa.AdditiveGaussianNoise(loc=0,
scale=scale,
per_channel=0.5)
])
images_aug = seq.augment_images(arr)
img = PIL.Image.fromarray(np.uint8(images_aug), 'RGB')
#SigmoidContrast, need further development
if self.contrast:
if random.random() > 0.5:
arr = np.asarray(img)
cutoff = random.uniform(0.45, 0.75)
gain = random.randint(5, 10)
seq = iaa.Sequential(
[iaa.SigmoidContrast(gain=gain, cutoff=cutoff)])
images_aug = seq.augment_images(arr)
img = PIL.Image.fromarray(np.uint8(images_aug), 'RGB')
#Sharpen, need further development
if self.sharpen:
if random.random() > 0.3:
arr = np.asarray(img)
alpha = random.uniform(0, 0.5)
seq = iaa.Sharpen(alpha=alpha)
images_aug = seq.augment_images(arr)
img = PIL.Image.fromarray(np.uint8(images_aug), 'RGB')
if self.vis_batch and self.data_aug:
vis_post_aug_img = copy.deepcopy(img)
tmp_path = os.path.join(visualization_tmp_path,
img_name[:-4] + "_post_augmentation.jpg")
visualize_and_save_to_local(vis_post_aug_img,
labels,
tmp_path,
box_color="green")
if self.vis_batch:
self.vis_counter += 1
if self.vis_counter > (self.vis_batch - 1):
sys.exit('Finished visualizing enough images. Exiting!')
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
labels[:, (1, 3)] /= self.width
labels[:, (2, 4)] /= self.height
img = torchvision.transforms.functional.to_tensor(img)
labels = F.pad(labels,
pad=[0, 0, 0, self.num_targets_per_image - len(labels)],
mode="constant")
if (labels < 0).sum() > 0:
raise Exception(f"labels for image {img_uri} have negative values")
return img_uri, img, labels
def __getitem__(self, index):
img_path = self.img_files[index]
label_path = self.label_files[index]
# if hasattr(self, 'imgs'):
# img = self.imgs[index] # BGR
img = cv2.imread(img_path) # BGR
assert img is not None, 'File Not Found ' + img_path
h, w, _ = img.shape
img, ratio, padw, padh = letterbox(
img, height=self.img_size) # 将每幅图resize到img_size
# Load labels
labels = []
if os.path.isfile(label_path):
with open(label_path, 'r') as file:
lines = file.read().splitlines(
) # 每一行的内容: class x_center y_center w h 比如 4 0.43 0.36 0.06 0.24，坐标都是归一化过的
x = np.array([x.split() for x in lines],
dtype=np.float32) # x: (box_num, 5)
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:, 1] = ratio * w * (
x[:, 1] - x[:, 3] / 2
) + padw # 因为图像resize了，所以labels中的坐标信息也要相对变化 TODO：理解的不是很透彻
labels[:, 2] = ratio * h * (x[:, 2] - x[:, 4] / 2) + padh
labels[:, 3] = ratio * w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:, 4] = ratio * h * (x[:, 2] + x[:, 4] / 2) + padh
print(labels)
# Augment image and labels
#if self.augment:
# img, labels = random_affine(img, labels, degrees=(-5, 5), translate=(0.10, 0.10), scale=(0.90, 1.10))
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5]) / self.img_size
# TODO
# if self.augment:
# # random left-right flip
# lr_flip = True
# if lr_flip and random.random() > 0.5:
# img = np.fliplr(img)
# if nL:
# labels[:, 1] = 1 - labels[:, 1]
#
# # random up-down flip
# ud_flip = False
# if ud_flip and random.random() > 0.5:
# img = np.flipud(img)
# if nL:
# labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros(
(nL, 6)) # clw note: maybe leave index 0 for batch_size dim
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Normalize
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img, dtype=np.float32) # uint8 to float32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
return torch.from_numpy(img), labels_out, img_path, (h, w)
def __getitem__(self, index):
img_path = self.img_files[index]
label_path = self.label_files[index]
# Load image
if hasattr(self, 'imgs'): # preloaded
img = self.imgs[index]
else:
img = cv2.imread(img_path) # BGR
assert img is not None, 'File Not Found ' + img_path
# Letterbox
h, w, *_ = img.shape
if self.pad_rectangular:
new_shape = self.batch_shapes[self.batch[index]]
img, ratio, padw, padh = letterbox(img,
new_shape=new_shape,
mode='rect')
else:
img, ratio, padw, padh = letterbox(img,
new_shape=self.img_size,
mode='square')
# Load labels
labels = []
if os.path.isfile(label_path):
# with open(label_path, 'r') as f:
# x = np.array([x.split() for x in f.read().splitlines()], dtype=np.float32)
x = self.labels[index]
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:, 1] = ratio * w * (x[:, 1] - x[:, 3] / 2) + padw
labels[:, 2] = ratio * h * (x[:, 2] - x[:, 4] / 2) + padh
labels[:, 3] = ratio * w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:, 4] = ratio * h * (x[:, 2] + x[:, 4] / 2) + padh
# Augment image and labels
if self.augment:
img, labels = random_affine(img,
labels,
degrees=(-5, 5),
translate=(0.10, 0.10),
scale=(0.90, 1.10))
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() > 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() > 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Normalize
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img, dtype=np.float32) # uint8 to float32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
return torch.from_numpy(img), labels_out, img_path, (h, w)
def detect(self):
frame_no = -1
# skip_no = 2
if self.output_file:
f = open(output_file, "w")
while self.vidCap.grab():
frame_no += 1
# skip frames every n frames
# if frame_no % skip_no != 0:
# continue
# start time
total_begin = time.time()
_, img = self.vidCap.retrieve()
# yolov3部分
yolo_begin = time.time()
bbox_xyxy, cls_conf, cls_ids = self.yolov3.predict(img)
# [x1,y1,x2,y2]
yolo_end = time.time()
# deepsort部分
ds_begin = time.time()
if bbox_xyxy is not None:
bbox_cxcywh = xyxy2xywh(bbox_xyxy)
outputs = self.deepsort.update(bbox_cxcywh, cls_conf, img)
if len(outputs) > 0:
# [x1,y1,x2,y2] id
bbox_xyxy = outputs[:, :4]
ids = outputs[:, -1]
img = draw_bboxes(img, bbox_xyxy, ids)
# frame,id,tlwh,1,-1,-1,-1
if self.output_file:
bbox_tlwh = xyxy2xywh(bbox_xyxy)
for i in range(len(bbox_tlwh)):
write_line = "%d,%d,%d,%d,%d,%d,1,-1,-1,-1\n" % (
frame_no + 1, outputs[i, -1],
int(bbox_tlwh[i][0]), int(bbox_tlwh[i][1]),
int(bbox_tlwh[i][2]), int(bbox_tlwh[i][3]))
f.write(write_line)
ds_end = time.time()
total_end = time.time()
if frame_no % 500 == 0:
print("frame:%04d|det:%.4f|deep sort:%.4f|total:%.4f|det p:%.2f%%|fps:%.2f" % (frame_no,
(yolo_end - yolo_begin),
(ds_end - ds_begin),
(total_end - total_begin),
((yolo_end - yolo_begin) * 100 / (
total_end - total_begin)),
(1 / (total_end - total_begin))))
if self.display is True:
cv2.imshow("Test", img)
cv2.waitKey(1)
if self.save_path:
self.output.write(img)
if self.output_file:
f.close()
def __getitem__(self, index):
if self.image_weights:
index = self.indices[index]
img_path = self.img_files[index]
label_path = self.label_files[index]
# Load image
img = self.imgs[index]
if img is None:
img = cv2.imread(img_path) # BGR
assert img is not None, 'File Not Found ' + img_path
if self.n < 1001:
self.imgs[index] = img # cache image into memory
# Augment colorspace
augment_hsv = True
if self.augment and augment_hsv:
# SV augmentation by 50%
fraction = 0.50 # must be < 1.0
img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV) # hue, sat, val
S = img_hsv[:, :, 1].astype(np.float32) # saturation
V = img_hsv[:, :, 2].astype(np.float32) # value
a = (random.random() * 2 - 1) * fraction + 1
b = (random.random() * 2 - 1) * fraction + 1
S *= a
V *= b
img_hsv[:, :, 1] = S if a < 1 else S.clip(None, 255)
img_hsv[:, :, 2] = V if b < 1 else V.clip(None, 255)
cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img)
# Letterbox
h, w, _ = img.shape
if self.rect:
shape = self.batch_shapes[self.batch[index]]
img, ratiow, ratioh, padw, padh = letterbox(img,
new_shape=shape,
mode='rect')
else:
shape = self.img_size
img, ratiow, ratioh, padw, padh = letterbox(img,
new_shape=shape,
mode='square')
# Load labels
labels = []
if os.path.isfile(label_path):
x = self.labels[index]
if x is None: # labels not preloaded
with open(label_path, 'r') as f:
x = np.array([x.split() for x in f.read().splitlines()],
dtype=np.float32)
self.labels[index] = x # save for next time
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:, 1] = ratiow * w * (x[:, 1] - x[:, 3] / 2) + padw
labels[:, 2] = ratioh * h * (x[:, 2] - x[:, 4] / 2) + padh
labels[:, 3] = ratiow * w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:, 4] = ratioh * h * (x[:, 2] + x[:, 4] / 2) + padh
# Augment image and labels
if self.augment:
img, labels = random_affine(img,
labels,
degrees=(-5, 5),
translate=(0.10, 0.10),
scale=(0.90, 1.10))
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() > 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() > 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Normalize
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img, dtype=np.float32) # uint8 to float32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
return torch.from_numpy(img), labels_out, img_path, (h, w)
def __next__(self):
self.count += 1
if self.count == self.nB:
raise StopIteration
ia = self.count * self.batch_size
ib = min((self.count + 1) * self.batch_size, self.nF)
multi_scale = False
if multi_scale and self.augment:
# Multi-Scale YOLO Training
height = random.choice(range(10, 20)) * 32 # 320 - 608 pixels
else:
# Fixed-Scale YOLO Training
height = self.height
img_all = []
labels_all = []
for index, files_index in enumerate(range(ia, ib)):
img_path = self.img_files[self.shuffled_vector[files_index]]
label_path = self.label_files[self.shuffled_vector[files_index]]
img = cv2.imread(img_path) # BGR
if img is None:
print('nooooooooooimages')
continue
augment_hsv = True
if self.augment and augment_hsv:
# SV augmentation by 50%
fraction = 0.50
img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
S = img_hsv[:, :, 1].astype(np.float32)
V = img_hsv[:, :, 2].astype(np.float32)
a = (random.random() * 2 - 1) * fraction + 1
S *= a
if a > 1:
np.clip(S, a_min=0, a_max=255, out=S)
a = (random.random() * 2 - 1) * fraction + 1
V *= a
if a > 1:
np.clip(V, a_min=0, a_max=255, out=V)
img_hsv[:, :, 1] = S.astype(np.uint8)
img_hsv[:, :, 2] = V.astype(np.uint8)
cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img)
h, w, _ = img.shape
img, ratio, padw, padh = resize_square(img,
height=height,
color=(127.5, 127.5, 127.5))
# Load labels
name_classes = load_classes(
'/Users/jx/Desktop/jjjjjxxxx/omr_yolo3/cfg/new_duration.names')
pitch_classes = load_classes(
'/Users/jx/Desktop/jjjjjxxxx/omr_yolo3/cfg/pitch.names')
if os.path.isfile(label_path):
labels0 = np.loadtxt(label_path,
dtype=np.float32).reshape(-1, 7)
# Normalized xywh to pixel xyxy format
labels = labels0.copy()
# labels[:, 1] = ratio * w * (labels0[:, 1] - labels0[:, 3] / 2) + padw
#labels[:, 2] = ratio * h * (labels0[:, 2] - labels0[:, 4] / 2) + padh
#labels[:, 3] = ratio * w * (labels0[:, 1] + labels0[:, 3] / 2) + padw
#labels[:, 4] = ratio * h * (labels0[:, 2] + labels0[:, 4] / 2) + padh
labels[:, 1] = ratio * labels0[:, 1] + padw
labels[:, 2] = ratio * labels0[:, 2] + padh
labels[:, 3] = ratio * labels0[:, 3] + padw
labels[:, 4] = ratio * labels[:, 4] + padh
durations = []
pitchs = []
# for i in labels0[:,5]:
# if float(i)==float(1):
# durations.append(9)
# elif str(i) not in name_classes and str(i)!='0.0':
# durations.append(6)
# else:
# for idx,j in enumerate(name_classes):
# if float(i)== float(j):
# durations.append(idx)
for i in labels0[:, 5]:
if str(int(i)) in name_classes:
durations.append(int(i))
else:
durations.append(10)
##超出音高范围或者没有音高
for i in labels0[:, 6]:
if str(int(i)) in pitch_classes:
pitchs.append(int(i))
elif int(i) > 15:
pitchs.append(15)
else:
pitchs.append(-5)
labels[:, 5] = durations
labels[:, 6] = pitchs
else:
labels = np.array([])
# Augment image and labels
if self.augment:
img, labels, M = random_affine(img,
labels,
degrees=(-3, 3),
translate=(0.1, 0.1),
scale=(0.9, 1.1))
plotFlag = False
if plotFlag:
import matplotlib.pyplot as plt
plt.figure(figsize=(10, 10)) if index == 0 else None
plt.subplot(4, 4, index + 1).imshow(img[:, :, ::-1])
plt.plot(labels[:, [1, 3, 3, 1, 1]].T,
labels[:, [2, 2, 4, 4, 2]].T, '.-')
plt.axis('off')
nL = len(labels)
if nL > 0:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5].copy()) / height
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip & (random.random() > 0.5):
img = np.fliplr(img)
if nL > 0:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip & (random.random() > 0.5):
img = np.flipud(img)
if nL > 0:
labels[:, 2] = 1 - labels[:, 2]
img_all.append(img)
labels_all.append(torch.from_numpy(labels))
# Normalize
assert len(img_all) != 0
img_all = np.stack(img_all)[:, :, :, ::-1].transpose(
0, 3, 1, 2) # BGR to RGB and cv2 to pytorch
img_all = np.ascontiguousarray(img_all, dtype=np.float32)
# img_all -= self.rgb_mean
# img_all /= self.rgb_std
img_all /= 255.0
return torch.from_numpy(img_all), labels_all
def __getitem__(self, index):
# 在训练中为false
if self.image_weights:
index = self.indices[index]
hyp = self.hyp
if self.mosaic:
# Load mosaic
img, labels = load_mosaic(self, index) # index是一个int类型
shapes = None
else:
# Load image
# h,w是经过调整之后的 其中有一个值等于img_size img是经过插值之后的图像(且是BGR格式) 其中一边等于img_size
img, (h0, w0), (h, w) = load_image(self, index)
# Letterbox
# shape存放的height 和 width
shape = self.batch_shapes[self.batch[
index]] if self.rect else self.img_size # final letterboxed shape
img, ratio, pad = letterbox(img,
shape,
auto=False,
scaleup=self.augment)
shapes = (h0, w0), (
(h / h0, w / w0), pad) # for COCO mAP rescaling
# Load labels
labels = []
# self.labels[index]表示index对应的图片中所有的gtbox []
x = self.labels[index]
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:,
1] = ratio[0] * w * (x[:, 1] -
x[:, 3] / 2) + pad[0] # pad width
labels[:,
2] = ratio[1] * h * (x[:, 2] -
x[:, 4] / 2) + pad[1] # pad height
labels[:, 3] = ratio[0] * w * (x[:, 1] + x[:, 3] / 2) + pad[0]
labels[:, 4] = ratio[1] * h * (x[:, 2] + x[:, 4] / 2) + pad[1]
if self.augment:
# Augment imagespace
if not self.mosaic:
img, labels = random_affine(img,
labels,
degrees=hyp['degrees'],
translate=hyp['translate'],
scale=hyp['scale'],
shear=hyp['shear'])
# Augment colorspace
augment_hsv(img,
hgain=hyp['hsv_h'],
sgain=hyp['hsv_s'],
vgain=hyp['hsv_v'])
# Apply cutouts
# if random.random() < 0.9:
# labels = cutout(img, labels)
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
# 这里的xyxy是未归一化的 xywh也是未归一化的
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() < 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
# [batch, cls, x, y, w, h]
labels_out = torch.zeros((nL, 6)) # nl
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Convert
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img)
# print(img.shape)
return torch.from_numpy(img), labels_out, self.img_files[index], shapes
def __getitem__(self, index):
if self.image_weights:
index = self.indices[index]
img_path = self.img_files[index]
label_path = self.label_files[index]
mosaic = True and self.augment # load 4 images at a time into a mosaic (only during training)
if mosaic:
# Load mosaic
img, labels = load_mosaic(self, index)
h, w, _ = img.shape
else:
# Load image
img = load_image(self, index)
# Letterbox
h, w, _ = img.shape
#Added code to perform warp affine to 640 X 384
'''
#cv2.imshow("Before padding: ", img0)
#cv2.waitKey(0)
inp_width = 640
inp_height = 384
c = np.array([1920 / 2., 1080/ 2.], dtype=np.float32)
s = 1920
trans_input = get_affine_transform(c, s, 0, [inp_width, inp_height])
inp_image = cv2.warpAffine(img0, trans_input, (inp_width, inp_height), flags=cv2.INTER_LINEAR)
#cv2.imshow('Warp Affine: ', inp_image)
#cv2.waitKey(0)
#img = letterbox(inp_image, new_shape=self.img_size)[0]
img = inp_image
'''
#End of added warp affine
if self.rect:
img, ratio, padw, padh = letterbox(
img, self.batch_shapes[self.batch[index]], mode='rect')
#Hard coded values
#img, ratio, padw, padh = letterbox(img, np.array([384, 640]), mode='rect')
else:
img, ratio, padw, padh = letterbox(img,
self.img_size,
mode='square')
'''
print("Ratio: ", ratio)
print("padw: ", padw)
print('padh: ', padh)
cv2.imshow('output', img)
if cv2.waitKey(0) == ord('q'):
exit()
'''
# Load labels
labels = []
if os.path.isfile(label_path):
x = self.labels[index]
if x is None: # labels not preloaded
with open(label_path, 'r') as f:
x = np.array(
[x.split() for x in f.read().splitlines()],
dtype=np.float32)
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:,
1] = ratio[0] * w * (x[:, 1] - x[:, 3] / 2) + padw
labels[:,
2] = ratio[1] * h * (x[:, 2] - x[:, 4] / 2) + padh
labels[:,
3] = ratio[0] * w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:,
4] = ratio[1] * h * (x[:, 2] + x[:, 4] / 2) + padh
if self.augment:
# Augment colorspace
augment_hsv(img,
hgain=self.hyp['hsv_h'],
sgain=self.hyp['hsv_s'],
vgain=self.hyp['hsv_v'])
# Augment imagespace
g = 0.0 if mosaic else 1.0 # do not augment mosaics
hyp = self.hyp
img, labels = random_affine(img,
labels,
degrees=hyp['degrees'] * g,
translate=hyp['translate'] * g,
scale=hyp['scale'] * g,
shear=hyp['shear'] * g)
# Apply cutouts
# if random.random() < 0.9:
# labels = cutout(img, labels)
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() < 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Normalize
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img, dtype=np.float32) # uint8 to float32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
return torch.from_numpy(img), labels_out, img_path, (h, w)
def get_data(self, img_path, label_path):
height = self.height
width = self.width
img = cv2.imread(img_path) # BGR
if img is None:
raise ValueError('File corrupt {}'.format(img_path))
augment_hsv = True
if self.augment and augment_hsv:
# SV augmentation by 50%
fraction = 0.50
img_hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
S = img_hsv[:, :, 1].astype(np.float32)
V = img_hsv[:, :, 2].astype(np.float32)
a = (random.random() * 2 - 1) * fraction + 1
S *= a
if a > 1:
np.clip(S, a_min=0, a_max=255, out=S)
a = (random.random() * 2 - 1) * fraction + 1
V *= a
if a > 1:
np.clip(V, a_min=0, a_max=255, out=V)
img_hsv[:, :, 1] = S.astype(np.uint8)
img_hsv[:, :, 2] = V.astype(np.uint8)
cv2.cvtColor(img_hsv, cv2.COLOR_HSV2BGR, dst=img)
h, w, _ = img.shape
img, ratio, padw, padh = letterbox(img, height=height, width=width)
# Load labels
if os.path.isfile(label_path):
labels0 = np.loadtxt(label_path, dtype=np.float32).reshape(-1, 6)
# Normalized xywh to pixel xyxy format
labels = labels0.copy()
labels[:,
2] = ratio * w * (labels0[:, 2] - labels0[:, 4] / 2) + padw
labels[:,
3] = ratio * h * (labels0[:, 3] - labels0[:, 5] / 2) + padh
labels[:,
4] = ratio * w * (labels0[:, 2] + labels0[:, 4] / 2) + padw
labels[:,
5] = ratio * h * (labels0[:, 3] + labels0[:, 5] / 2) + padh
else:
labels = np.array([])
# Augment image and labels
if self.augment:
img, labels, M = random_affine(img,
labels,
degrees=(-5, 5),
translate=(0.10, 0.10),
scale=(0.50, 1.20))
plotFlag = False
if plotFlag:
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
plt.figure(figsize=(50, 50))
plt.imshow(img[:, :, ::-1])
plt.plot(labels[:, [1, 3, 3, 1, 1]].T,
labels[:, [2, 2, 4, 4, 2]].T, '.-')
plt.axis('off')
plt.savefig('test.jpg')
time.sleep(10)
nL = len(labels)
if nL > 0:
# convert xyxy to xywh
labels[:, 2:6] = xyxy2xywh(labels[:, 2:6].copy()) # / height
labels[:, 2] /= width
labels[:, 3] /= height
labels[:, 4] /= width
labels[:, 5] /= height
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip & (random.random() > 0.5):
img = np.fliplr(img)
if nL > 0:
labels[:, 2] = 1 - labels[:, 2]
img = np.ascontiguousarray(img[:, :, ::-1]) # BGR to RGB
if self.transforms is not None:
img = self.transforms(img)
return img, labels, img_path, (h, w)
def __getitem__(self, index):
if self.image_weights:
index = self.indices[index]
img_path = self.img_files[index]
label_path = self.label_files[index]
mosaic = True and self.augment # load 4 images at a time into a mosaic (only during training)
if mosaic:
# Load mosaic
img, labels = load_mosaic(self, index)
h, w, _ = img.shape
else:
# Load image
img = load_image(self, index)
# Letterbox
h, w, _ = img.shape
if self.rect:
img, ratio, padw, padh = letterbox(
img, self.batch_shapes[self.batch[index]], mode='rect')
else:
img, ratio, padw, padh = letterbox(img,
self.img_size,
mode='square')
# Load labels
labels = []
if os.path.isfile(label_path):
x = self.labels[index]
if x is None: # labels not preloaded
with open(label_path, 'r') as f:
x = np.array(
[x.split() for x in f.read().splitlines()],
dtype=np.float32)
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:,
1] = ratio[0] * w * (x[:, 1] - x[:, 3] / 2) + padw
labels[:,
2] = ratio[1] * h * (x[:, 2] - x[:, 4] / 2) + padh
labels[:,
3] = ratio[0] * w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:,
4] = ratio[1] * h * (x[:, 2] + x[:, 4] / 2) + padh
if self.augment:
# Augment imagespace
g = 0.0 if mosaic else 1.0 # do not augment mosaics
hyp = self.hyp
img, labels = random_affine(img,
labels,
degrees=hyp['degrees'] * g,
translate=hyp['translate'] * g,
scale=hyp['scale'] * g,
shear=hyp['shear'] * g)
# Apply cutouts
# if random.random() < 0.9:
# labels = cutout(img, labels)
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() < 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Normalize
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img, dtype=np.float32) # uint8 to float32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
return torch.from_numpy(img), labels_out, img_path, (h, w)
def __getitem__(self, index):
if self.image_weights:
index = self.indices[index]
hyp = self.hyp
if self.mosaic:
# Load mosaic
img, labels = load_mosaic(self, index)
shapes = None
# MixUp https://arxiv.org/pdf/1710.09412.pdf
# if random.random() < 0.5:
# img2, labels2 = load_mosaic(self, random.randint(0, len(self.labels) - 1))
# r = np.random.beta(0.3, 0.3) # mixup ratio, alpha=beta=0.3
# img = (img * r + img2 * (1 - r)).astype(np.uint8)
# labels = np.concatenate((labels, labels2), 0)
else:
# Load image
img, (h0, w0), (h, w) = load_image(self, index)
# Letterbox
shape = self.batch_shapes[self.batch[
index]] if self.rect else self.img_size # final letterboxed shape
img, ratio, pad = letterbox(img,
shape,
auto=False,
scaleup=self.augment)
shapes = (h0, w0), (
(h / h0, w / w0), pad) # for COCO mAP rescaling
# Load labels
labels = []
x = self.labels[index]
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:,
1] = ratio[0] * w * (x[:, 1] -
x[:, 3] / 2) + pad[0] # pad width
labels[:,
2] = ratio[1] * h * (x[:, 2] -
x[:, 4] / 2) + pad[1] # pad height
labels[:, 3] = ratio[0] * w * (x[:, 1] + x[:, 3] / 2) + pad[0]
labels[:, 4] = ratio[1] * h * (x[:, 2] + x[:, 4] / 2) + pad[1]
if self.augment:
# Augment imagespace
if not self.mosaic:
img, labels = random_affine(img,
labels,
degrees=hyp['degrees'],
translate=hyp['translate'],
scale=hyp['scale'],
shear=hyp['shear'])
# Augment colorspace
augment_hsv(img,
hgain=hyp['hsv_h'],
sgain=hyp['hsv_s'],
vgain=hyp['hsv_v'])
# Apply cutouts
# if random.random() < 0.9:
# labels = cutout(img, labels)
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() < 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Convert
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img)
return torch.from_numpy(img), labels_out, self.img_files[index], shapes
def augment_collection(self, index):
if self.image_weights:
index = self.indices[index]
img_path = self.img_files[index]
label_path = self.label_files[index]
mosaic = True and self.augment # load 4 images at a time into a mosaic (only during training)
if mosaic and random.random() < 0.5: # modify: add random to mosaic
# Load mosaic
img, labels = load_mosaic(self, index)
h, w = img.shape[:2]
else:
# Load image
img = load_image(self, index)
# Letterbox
h, w = img.shape[:2]
if self.rect:
img, ratio, padw, padh = letterbox(
img, self.batch_shapes[self.batch[index]], mode='rect')
else:
img, ratio, padw, padh = letterbox(img,
self.img_size,
mode='square')
# Load labels
labels = []
if os.path.isfile(label_path):
x = self.labels[index]
if x is None: # labels not preloaded
with open(label_path, 'r') as f:
x = np.array(
[x.split() for x in f.read().splitlines()],
dtype=np.float32)
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:,
1] = ratio[0] * w * (x[:, 1] - x[:, 3] / 2) + padw
labels[:,
2] = ratio[1] * h * (x[:, 2] - x[:, 4] / 2) + padh
labels[:,
3] = ratio[0] * w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:,
4] = ratio[1] * h * (x[:, 2] + x[:, 4] / 2) + padh
if self.augment:
# Augment colorspace
augment_hsv(img,
hgain=self.hyp['hsv_h'],
sgain=self.hyp['hsv_s'],
vgain=self.hyp['hsv_v'])
# Augment imagespace
g = 0.0 if mosaic else 1.0 # do not augment mosaics
g = 1
hyp = self.hyp
img, labels = random_affine(img,
labels,
degrees=hyp['degrees'] * g,
translate=hyp['translate'] * g,
scale=hyp['scale'] * g,
shear=hyp['shear'] * g)
# Apply cutouts
# if random.random() < 0.9:
# labels = cutout(img, labels)
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False # acitve for topview
if ud_flip and random.random() < 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
# labels_out = torch.zeros((nL, 6)) # todo: add gt_score to labels_out(i.e. targets)
labels_out = np.zeros(
(nL, 7),
dtype='float32') # add gt_score to labels_out(i.e. targets)
if nL:
# labels_out[:, 1:] = torch.from_numpy(labels)
labels_out[:, 1] = (labels[:, 0]) # cls
labels_out[:, 2] = 1 # gt_score
labels_out[:, 3:] = (labels[:, 1:]) # [x y w h]
# Normalize
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img, dtype=np.float32) # uint8 to float32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if norm_with_mean_std:
img[0] = (img[0] - 0.485) / 0.229
img[1] = (img[1] - 0.456) / 0.224
img[2] = (img[2] - 0.406) / 0.225
return img, labels_out, img_path, (h, w)
def __getitem__(self, index):
if self.image_weights:
index = self.indices[index]
img_path = self.img_files[index]
label_path = self.label_files[index]
hyp = self.hyp
mosaic = True and self.augment # load 4 images at a time into a mosaic (only during training)
if mosaic:
# Load mosaic
img, labels = load_mosaic(self, index)
h, w = img.shape[:2]
ratio, pad = None, None
else:
# Load image
img = load_image(self, index)
# Letterbox
h, w = img.shape[:2]
shape = self.batch_shapes[self.batch[
index]] if self.rect else self.img_size # final letterboxed shape
img, ratio, pad = letterbox(img,
shape,
auto=False,
scaleup=self.augment)
# Load labels
labels = []
if os.path.isfile(label_path):
x = self.labels[index]
if x is None: # labels not preloaded
with open(label_path, 'r') as f:
x = np.array(
[x.split() for x in f.read().splitlines()],
dtype=np.float32)
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:,
1] = ratio[0] * w * (x[:, 1] - x[:, 3] / 2) + pad[
0] # pad width
labels[:,
2] = ratio[1] * h * (x[:, 2] - x[:, 4] / 2) + pad[
1] # pad height
labels[:,
3] = ratio[0] * w * (x[:, 1] + x[:, 3] / 2) + pad[0]
labels[:,
4] = ratio[1] * h * (x[:, 2] + x[:, 4] / 2) + pad[1]
if self.augment:
# Augment imagespace
if not mosaic:
img, labels = random_affine(img,
labels,
degrees=hyp['degrees'],
translate=hyp['translate'],
scale=hyp['scale'],
shear=hyp['shear'])
# Augment colorspace
augment_hsv(img,
hgain=hyp['hsv_h'],
sgain=hyp['hsv_s'],
vgain=hyp['hsv_v'])
# Apply cutouts
# if random.random() < 0.9:
# labels = cutout(img, labels)
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() < 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Convert
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img)
return torch.from_numpy(img), labels_out, img_path, ((h, w), (ratio,
pad))
def __getitem__(self, index):
if self.image_weights:
index = self.indices[index]
hyp = self.hyp
if self.mosaic:
# Load mosaic
img, labels = load_mosaic(self, index)
shapes = None
else:
# Load image
img, (h0, w0), (h, w) = load_image(self, index)
# Letterbox
shape = self.batch_shapes[self.batch[
index]] if self.rect else self.img_size # final letterboxed shape
img, ratio, pad = letterbox(img,
shape,
auto=False,
scaleup=self.augment)
shapes = (h0, w0), (
(h / h0, w / w0), pad) # for COCO mAP rescaling
# Load labels
labels = []
x = self.labels[index]
if x is not None and x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:,
1] = ratio[0] * w * (x[:, 1] -
x[:, 3] / 2) + pad[0] # pad width
labels[:,
2] = ratio[1] * h * (x[:, 2] -
x[:, 4] / 2) + pad[1] # pad height
labels[:, 3] = ratio[0] * w * (x[:, 1] + x[:, 3] / 2) + pad[0]
labels[:, 4] = ratio[1] * h * (x[:, 2] + x[:, 4] / 2) + pad[1]
#labels[:, 5:] = x[:, 5:] ###6
for i in range(5, 21, 2): ###12
labels[:, i] = ratio[0] * w * x[:, i] + pad[0] ###12
labels[:,
i + 1] = ratio[1] * h * x[:, i + 1] + pad[1] ###12
if self.augment:
# Augment imagespace
if not self.mosaic:
img, labels = random_affine(img,
labels,
degrees=hyp['degrees'],
translate=hyp['translate'],
scale=hyp['scale'],
shear=hyp['shear'])
# Augment colorspace
augment_hsv(img,
hgain=hyp['hsv_h'],
sgain=hyp['hsv_s'],
vgain=hyp['hsv_v'])
# Apply cutouts
# if random.random() < 0.9:
# labels = cutout(img, labels)
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
for i in range(5, 21, 2): ###12
labels[:, i + 1] /= img.shape[0] ###12
labels[:, i] /= img.shape[1] ###12
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
for i in range(5, 20, 2): ###12
labels[:, i] = 1 - labels[:, i] ###12
# random up-down flip
ud_flip = False
if ud_flip and random.random() < 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
for i in range(6, 21, 2): ###12
labels[:, i] = 1 - labels[:, i] ###12
labels_out = torch.zeros((nL, 6 + 16)) ###6
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Convert
img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = np.ascontiguousarray(img)
return torch.from_numpy(img), labels_out, self.img_files[index], shapes
def __getitem__(self, index): # 重写 Dataset父类的方法
hyp = self.hyp
if self.mosaic:
# load mosaic
img, labels = load_mosaic(self, index)
shapes = None
else:
# load image
img, (h0, w0), (h, w) = load_image(self, index)
# letterbox
shape = self.batch_shapes[self.batch[
index]] if self.rect else self.img_size # final letterboxed shape
img, ratio, pad = letterbox(img,
shape,
auto=False,
scale_up=self.augment)
shapes = (h0, w0), (
(h / h0, w / w0), pad) # for COCO mAP rescaling
# load labels
labels = []
x = self.labels[index]
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy() # label: class, x, y, w, h
labels[:,
1] = ratio[0] * w * (x[:, 1] -
x[:, 3] / 2) + pad[0] # pad width
labels[:,
2] = ratio[1] * h * (x[:, 2] -
x[:, 4] / 2) + pad[1] # pad height
labels[:, 3] = ratio[0] * w * (x[:, 1] + x[:, 3] / 2) + pad[0]
labels[:, 4] = ratio[1] * h * (x[:, 2] + x[:, 4] / 2) + pad[1]
if self.augment:
# Augment imagespace
if not self.mosaic:
img, labels = random_affine(img,
labels,
degrees=hyp["degrees"],
translate=hyp["translate"],
scale=hyp["scale"],
shear=hyp["shear"])
# Augment colorspace
augment_hsv(img,
h_gain=hyp["hsv_h"],
s_gain=hyp["hsv_s"],
v_gain=hyp["hsv_v"])
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0-1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1] # 1 - x_center
# random up-down flip
ud_flip = False
if ud_flip and random.random() < 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2] # 1 - y_center
labels_out = torch.zeros((nL, 6)) # nL: number of labels
if nL:
# labels_out[:, 0] = index
labels_out[:, 1:] = torch.from_numpy(labels)
# Convert BGR to RGB, and HWC to CHW(3x512x512)
img = img[:, :, ::-1].transpose(2, 0, 1)
img = np.ascontiguousarray(img)
return torch.from_numpy(
img), labels_out, self.img_files[index], shapes, index
def __getitem__(self, index):
if self.image_weights:
index = self.indices[index]
img_path = self.img_files[index]
label_path = self.label_files[index]
mosaic = True and self.augment # load 4 images at a time into a mosaic (only during training)
if mosaic:
# Load mosaic
img, labels = load_mosaic(self, index)
h, w = img.shape[:2]
else:
# Load image
img = load_image(self, index)
# Letterbox
h, w = img.shape[:2]
if self.rect:
img, ratio, padw, padh = letterbox(
img, self.batch_shapes[self.batch[index]], mode='rect')
else:
img, ratio, padw, padh = letterbox(img,
self.img_size,
mode='square')
# Load labels
labels = []
if os.path.isfile(label_path):
x = self.labels[index]
if x is None: # labels not preloaded
with open(label_path, 'r') as f:
x = np.array(
[x.split() for x in f.read().splitlines()],
dtype=np.float32)
if x.size > 0:
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:,
1] = ratio[0] * w * (x[:, 1] - x[:, 3] / 2) + padw
labels[:,
2] = ratio[1] * h * (x[:, 2] - x[:, 4] / 2) + padh
labels[:,
3] = ratio[0] * w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:,
4] = ratio[1] * h * (x[:, 2] + x[:, 4] / 2) + padh
if self.augment or self.augment is False: #notice test need't augment,so the channel is bgr
# # Augment colorspace
# augment_hsv(img, hgain=self.hyp['hsv_h'], sgain=self.hyp['hsv_s'], vgain=self.hyp['hsv_v'])
## cv2.imshow("xxxx",img)
## cv2.waitKey(0)
# # Augment imagespace
# g = 0.0 if mosaic else 1.0 # do not augment mosaics
# hyp = self.hyp
# img, labels = random_affine(img, labels,
# degrees=hyp['degrees'] * g,
# translate=hyp['translate'] * g,
# scale=hyp['scale'] * g,
# shear= hyp['shear'] *g)
# Augment colorspace
img = img[:, :, (2, 1, 0)] #bgr(cv2) to rgb(plt)
p = [
iaa.Multiply([1, 2.5, 0.5, 1.5]),
iaa.SigmoidContrast(gain=10, cutoff=[0.75, 1, 0.5]),
# iaa.SigmoidContrast(gain=3,cutoff=[0,1],per_channel=0.9)
]
sequence_iaa = iaa.Sequential([random.choice(p)])
img = sequence_iaa.augment_image(img)
# Augment imagespace
g = 0.0 if mosaic else 1.0 # do not augment mosaics
hyp = self.hyp
img, labels = random_affine(img,
labels,
degrees=0,
translate=0,
scale=0,
shear=0)
# cv2.imshow("xxxx",img)
# cv2.waitKey(0)
# print(hyp['shear'] * g)
# from matplotlib import pyplot as plt
# plt.imshow(img)
# plt.xticks([]), plt.yticks([]) # to hide tick values on X and Y axis
# plt.show()
# Apply cutouts
# if random.random() < 0.9:
# labels = cutout(img, labels)
nL = len(labels) # number of labels
if nL:
# convert xyxy to xywh
labels[:, 1:5] = xyxy2xywh(labels[:, 1:5])
# Normalize coordinates 0 - 1
labels[:, [2, 4]] /= img.shape[0] # height
labels[:, [1, 3]] /= img.shape[1] # width
if self.augment:
# random left-right flip
lr_flip = True
if lr_flip and random.random() < 0.5:
img = np.fliplr(img)
if nL:
labels[:, 1] = 1 - labels[:, 1]
# random up-down flip
ud_flip = False
if ud_flip and random.random() < 0.5:
img = np.flipud(img)
if nL:
labels[:, 2] = 1 - labels[:, 2]
labels_out = torch.zeros((nL, 6))
if nL:
labels_out[:, 1:] = torch.from_numpy(labels)
# Normalize
# img = img[:, :, ::-1].transpose(2, 0, 1) # BGR to RGB, to 3x416x416
img = img[:, :, :].transpose(2, 0, 1) # RGB, to 3x416x416
img = np.ascontiguousarray(img, dtype=np.float32) # uint8 to float32
img /= 255.0 # 0 - 255 to 0.0 - 1.0
return torch.from_numpy(img), labels_out, img_path, (h, w)
