1.Mixup
主要是思想:将同一批次中的图进行两两融合
特点:
1.融合两张图: alpha * img1 +(1-alpha)*img2
2.只在同一个批次中进行: [1,2,3,4] 同打散的 [2,1,4,3]
3.标签也进行融合: l = alpha*loss(output, label1) + (1-alpha)*loss(output, lable2)
4.主要是针对分类
效果:
![](https://i-blog.csdnimg.cn/blog_migrate/7f52f7502fee7eafcbfe97ae0f2c410a.png)
for i,(images,target) in enumerate(train_loader):
# 1.input output
images = images.cuda()
target = torch.from_numpy(np.array(target)).float().cuda()
# 2.mixup
# beta distribution: change shape by setting different shape:
# https://www.sciencedirect.com/topics/computer-science/beta-distribution
alpha=config.alpha # by default we could set to 1.0 (uniform dist.)
ratio = np.random.beta(alpha, alpha) # E.g. lam = 0.432
# randomly return a permutation. E.g torch.randperm(5)=> tensor([2, 3, 0, 4, 1])/random permutation
index = torch.randperm(images.size(0)).cuda()
inputs = ratio * images + (1 - ratio) * images[index] # [0,1,2,3,4] + [2,3,0,4,1]
# image mixed up, so do the targets
# 2 种实现:还一种:targets = ratio * target + (1 - ratio) * target[index]
# 后面只用一个criterion就行了
targets_a, targets_b = target, target[index]
outputs = model(inputs)
# do not forget the compound/mixed up criterion
loss = ratio * criterion(outputs, targets_a) + (1 - ratio) * criterion(outputs, targets_b) #融合损失
# 3.backward
optimizer.zero_grad() # reset gradient
loss.backward()
optimizer.step() # update parameters of net
2.Cutout
主要思想:扣除图片中的一些块
特点:
1.首先归一化
2.扣除一些块用灰色填充
效果:1.随机选一块扣除 2.用网格的方式扣除
![](https://i-blog.csdnimg.cn/blog_migrate/2bc1b77d160f0d937501a6b12330034f.png)
伪代码:随机扣除n_holes个、大小为length的块
''' Cutout '''
# 原始用的是cifar10 (size=32, padding=4)
# 怕补0影响结果,所以先要normalize, 再cutout
normalize = transforms.Normalize(mean=[x / 255.0 for x in [125.3, 123.0, 113.9]], # (rgb order)
std=[x / 255.0 for x in [63.0, 62.1, 66.7]])
train_transform = transforms.Compose([])
train_transform.transforms.append(transforms.ToTensor()) # to tensor: [0.0, 1.0]
train_transform.transforms.append(normalize)
# 注意了!
train_transform.transforms.append(Cutout(n_holes=1, length=16)) # by default
class Cutout(object):
"""Randomly mask out one or more patches from an image.
Args:
n_holes (int): Number of patches to cut out of each image.
length (int): The length (in pixels) of each square patch.
"""
def __init__(self, n_holes, length):
self.n_holes = n_holes
self.length = length
def __call__(self, img):
"""
Args:
img (Tensor): Tensor image of size (C, H, W).
Returns:
Tensor: Image with n_holes of dimension length x length cut out of it.
"""
h = img.size(1)
w = img.size(2)
mask = np.ones((h, w), np.float32)
for n in range(self.n_holes):
y = np.random.randint(h)
x = np.random.randint(w)
y1 = np.clip(y - self.length // 2, 0, h) # np.clip此处:将范围限定在[0, h]
y2 = np.clip(y + self.length // 2, 0, h)
x1 = np.clip(x - self.length // 2, 0, w)
x2 = np.clip(x + self.length // 2, 0, w)
mask[y1: y2, x1: x2] = 0.
mask = torch.from_numpy(mask)
mask = mask.expand_as(img)
img = img * mask
return img
3.CutMix
主要思想:从另外一幅图中扣一部分贴过来
特点:
1. mixup + cutout
2.只合并两张图
效果:
![](https://i-blog.csdnimg.cn/blog_migrate/77475e1a26c6ad3d864d25b0bc4c0509.png)
![](https://i-blog.csdnimg.cn/blog_migrate/620a5c4ec180386ce442b42a44a17214.png)
伪代码:
''' cutmix '''
def rand_bbox(size, lam):
W = size[2]
H = size[3]
cut_rat = np.sqrt(1. - lam)
cut_w = np.int(W * cut_rat)
cut_h = np.int(H * cut_rat)
# uniform
cx = np.random.randint(W)
cy = np.random.randint(H)
bbx1 = np.clip(cx - cut_w // 2, 0, W)
bby1 = np.clip(cy - cut_h // 2, 0, H)
bbx2 = np.clip(cx + cut_w // 2, 0, W)
bby2 = np.clip(cy + cut_h // 2, 0, H)
return bbx1, bby1, bbx2, bby2
# generate mixed sample
lam = np.random.beta(args.beta, args.beta) # 1.0, we use uniform dist.
rand_index = torch.randperm(input.size()[0]).cuda()
target_a = target
target_b = target[rand_index]
bbx1, bby1, bbx2, bby2 = rand_bbox(input.size(), lam)
input[:, :, bbx1:bbx2, bby1:bby2] = input[rand_index, :, bbx1:bbx2, bby1:bby2]
# adjust lambda to exactly match pixel ratio
lam = 1 - ((bbx2 - bbx1) * (bby2 - bby1) / (input.size()[-1] * input.size()[-2]))
# compute output
output = model(input)
loss = criterion(output, target_a) * lam + criterion(output, target_b) * (1. - lam) #计算损失的时候也是融合
4.mosaic
主要思想:就是将四张图拼在一起
特点:增强背景的复杂度
效果:
对于分类来说可以根据大小作为权重融合损失
对于标注来说相当于一张图里面有四个目标
![](https://i-blog.csdnimg.cn/blog_migrate/d02666c68e2bbdd820548faeb23391e0.png)
def load_mosaic(self, index):
# loads images in a mosaic
labels4 = []
s = self.img_size
xc, yc = [int(random.uniform(s * 0.5, s * 1.5)) for _ in range(2)] # mosaic center x, y
indices = [index] + [random.randint(0, len(self.labels) - 1) for _ in range(3)] # 3 additional image indices
# 在可允许的范围之内,随机抽取4个indices (batchsize=16, [1, 0, 2, 4], [2, 0, 1, 4])
for i, index in enumerate(indices):
# Load image
img, _, (h, w) = load_image(self, index)
# place img in img4
if i == 0: # top left
# 把新图像先设置成原来的4倍,到时候再resize回去,114是gray
img4 = np.full((s * 2, s * 2, img.shape[2]), 114, dtype=np.uint8) # base image with 4 tiles
x1a, y1a, x2a, y2a = max(xc - w, 0), max(yc - h, 0), xc, yc # xmin, ymin, xmax, ymax (new/large image)
x1b, y1b, x2b, y2b = w - (x2a - x1a), h - (y2a - y1a), w, h # xmin, ymin, xmax, ymax (original/small image)
# 回看ppt讲解
elif i == 1: # top right
x1a, y1a, x2a, y2a = xc, max(yc - h, 0), min(xc + w, s * 2), yc
x1b, y1b, x2b, y2b = 0, h - (y2a - y1a), min(w, x2a - x1a), h
elif i == 2: # bottom left
x1a, y1a, x2a, y2a = max(xc - w, 0), yc, xc, min(s * 2, yc + h)
x1b, y1b, x2b, y2b = w - (x2a - x1a), 0, max(xc, w), min(y2a - y1a, h)
elif i == 3: # bottom right
x1a, y1a, x2a, y2a = xc, yc, min(xc + w, s * 2), min(s * 2, yc + h)
x1b, y1b, x2b, y2b = 0, 0, min(w, x2a - x1a), min(y2a - y1a, h)
img4[y1a:y2a, x1a:x2a] = img[y1b:y2b, x1b:x2b] # img4[ymin:ymax, xmin:xmax]
padw = x1a - x1b # 有时边上还是灰的
padh = y1a - y1b
# Labels
x = self.labels[index]
labels = x.copy()
if x.size > 0: # Normalized xywh to pixel xyxy format
# 此时x是0-1,同时,label是[bbox_xc, bbox_yc, bbox_w, bbox_c]
labels[:, 1] = w * (x[:, 1] - x[:, 3] / 2) + padw
labels[:, 2] = h * (x[:, 2] - x[:, 4] / 2) + padh
labels[:, 3] = w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:, 4] = h * (x[:, 2] + x[:, 4] / 2) + padh
labels4.append(labels)
# Concat/clip labels
if len(labels4):
# a = np.array([[1, 2], [3, 4]])
# c = np.concatenate(a, axis=0)
# c: [1, 2, 3, 4]
labels4 = np.concatenate(labels4, 0) # 0是dimension
# np.clip(labels4[:, 1:] - s / 2, 0, s, out=labels4[:, 1:]) # use with center crop
np.clip(labels4[:, 1:], 0, 2 * s, out=labels4[:, 1:]) # use with random_affine
# Augment
# img4 = img4[s // 2: int(s * 1.5), s // 2:int(s * 1.5)] # center crop (WARNING, requires box pruning)
img4, labels4 = random_affine(img4, labels4,
degrees=self.hyp['degrees'],
translate=self.hyp['translate'],
scale=self.hyp['scale'],
shear=self.hyp['shear'],
border=-s // 2) # border to remove
return img4, labels4