1. 图像读取
1.1 Pillow
Pillow是Python图像处理函数库(PIL)的一个分支,提供了常见的图像处理和处理的操作。
from PIL import Image
im =Image.open('IMG/Task02/Pillow读取原图.png')
im
from PIL import Image,ImageFilter
im = Image.open('IMG/Task02/Pillow读取原图.png')
im2 = im.filter(ImageFilter.BLUR)
im2.save('blur.jpg','jpeg') #模糊处理
im2
from PIL import Image
im = Image.open('IMG/Task02/Pillow读取原图.png')
w,h = im.size
im.thumbnail((w//2,h//2)) # 缩放
im.save('thumbnail.jpg','jpeg')
im
1.2 Opencv
import cv2
img = cv2.imread('IMG/Task02/cat.png')
img = cv2.cvtColor(img,cv2.COLOR_BGR2RGB) # 颜色通道默认为BGR,转为RGB
cv2.imshow('cat',img)
cv2.waitKey()
import cv2
img = cv2.imread('IMG/Task02/cat.png')
img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY) # 灰度图
cv2.imshow('gray',img)
cv2.waitKey()
import cv2
img = cv2.imread('IMG/Task02/cat.png')
img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
edges = cv2.Canny(img,30,70)
cv2.imshow('canny',edges) # 边缘检测
cv2.waitKey()
cv2.imwrite('canny.jpg',edges)
2. 数据扩增
数据扩增的目的在于提升样本数目,避免过拟合,提高模型泛化能力。常见的方法有:
transforms.CenterCrop(对图片中心进行裁剪),
transforms.ColorJitter(变化颜色对比度,饱和度和零度),
transforms.FiveCrop(对图像四个角和中心进行裁剪以得到五分图像),
transforms.Grayscale(灰度变换),
transforms.Pad(像素填充),
transforms.RandomAffine(随机仿射变化),
transforms.RandomCrop(随机区域裁剪),
transforms.RandomHorizontalFlip(随机水平翻转),
transforms.RandomRotation(随机旋转),
transforms.RandomVerticalFlip(随机垂直旋转)
3. 读取数据
在Pytorch中数据通过Dataset进行封装,并通过DataLoder进行并行读取。所以只需重载一下数据逻辑就可以完成数据的读取。
Dataset:对数据集的封装,提供索引方式对数据样本进行读取。
DataLoder:对Dataset进行封装,提供批量的迭代读取。
import os, sys, glob, shutil, json
import cv2
from PIL import Image
import numpy as np
import torch
from torch.utils.data.dataset import Dataset
import torchvision.transforms as transforms
class SVHNDataset(Dataset):
def __init__(self, img_path, img_label, transform=None):
self.img_path = img_path
self.img_label = img_label
if transform is not None:
self.transform = transform
else:
self.transform = None
def __getitem__(self, index):
img = Image.open(self.img_path[index]).convert('RGB')
if self.transform is not None:
img = self.transform(img)
# 原始SVHN中类别10为数字0
lbl = np.array(self.img_label[index], dtype=np.int)
lbl = list(lbl) + (5 - len(lbl)) * [10]
return img, torch.from_numpy(np.array(lbl[:5]))
def __len__(self):
return len(self.img_path)
train_path = glob.glob('input/train/*.png')
train_path.sort()
train_json = json.load(open('input/train.json'))
train_label = [train_json[x]['label'] for x in train_json]
train_loader = torch.utils.data.DataLoader(
SVHNDataset(train_path, train_label,
transforms.Compose([
transforms.Resize((64, 128)), # 缩放到固定尺寸
transforms.ColorJitter(0.3, 0.3, 0.2), # 颜色变换
transforms.RandomRotation(5), # 随机旋转
transforms.ToTensor(), # 将图片转为tensor
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) # 像素归一化
])),
batch_size=10, # 每批样本个数
shuffle=False, # 是否打乱顺序
num_workers=0, # 读取的线程个数
)
for data in train_loader:
break
数据按照批次获取,每批次调用Dataset读取单个样本进行拼接。此时data的格式为:
torch.Size([10, 3, 64, 128]), torch.Size([10, 6])