本文介绍利用MNIST数据集构建多数字识别模型的过程。先通过预处理MNIST数据，拼接生成含多个数字的训练集和测试集；接着安装PaddleOCR及依赖，下载预训练模型；然后训练模型并导出；最后采样测试图片，用导出的模型进行识别测试。
引入
传统的基于 MNIST 数据集的手写数字识别模型只能识别单个数字但实际使用环境中，多数字识别才是更加常见的情况本次就使用 MNIST 数据集，通过拼接数据的方式，实现多数字识别模型构建数据集
拼接采样数据集In [ ]%cd ~!mkdir dataset !mkdir dataset/train!mkdir dataset/testimport cv2import randomimport numpy as npfrom tqdm import tqdmfrom paddle.vision.datasets import MNIST# 加载数据集mnist_train = MNIST(mode='train', backend='cv2')mnist_test = MNIST(mode='test', backend='cv2')# 数据集预处理datas_train = {}for i in range(len(mnist_train)): sample = mnist_train[i] x, y = sample[0], sample[1] _sum = np.sum(x, axis=0) _where = np.where(_sum > 0) x = 255 - x[:, _where[0][0]: _where[0][-1]+1] if str(y[0]) in datas_train: datas_train[str(y[0])].append(x) else: datas_train[str(y[0])] = [x]datas_test = {}for i in range(len(mnist_test)): sample = mnist_test[i] x, y = sample[0], sample[1] _sum = np.sum(x, axis=0) _where = np.where(_sum > 0) x = 255 - x[:, _where[0][0]: _where[0][-1]+1] if str(y[0]) in datas_test: datas_test[str(y[0])].append(x) else: datas_test[str(y[0])] = [x]# 图片拼接采样datas_train_list = []for num in tqdm(range(0, 999)): for _ in range(1000): imgs = [255 - np.zeros((28, np.random.randint(10)))] for word in str(num): index = np.random.randint(0, len(datas_train[word])) imgs.append(datas_train[word][index]) imgs.append(255 - np.zeros((28, np.random.randint(10)))) img = np.concatenate(imgs, 1) cv2.imwrite('dataset/train/%03d_%04d.webp' % (num, _), img) datas_train_list.append('train/%03d_%04d.webp\t%d\n' % (num, _, num))datas_test_list = []for num in tqdm(range(0, 999)): for _ in range(50): imgs = [255 - np.zeros((28, np.random.randint(10)))] for word in str(num): index = np.random.randint(0, len(datas_test[word])) imgs.append(datas_test[word][index]) imgs.append(255 - np.zeros((28, np.random.randint(10)))) img = np.concatenate(imgs, 1) cv2.imwrite('dataset/test/%03d_%04d.webp' % (num, _), img) datas_test_list.append('test/%03d_%04d.webp\t%d\n' % (num, _, num))# 数据列表生成with open('dataset/train.txt', 'w') as f: for line in datas_train_list: f.write(line)with open('dataset/test.txt', 'w') as f: for line in datas_test_list: f.write(line)登录后复制 数据样例展示
安装 PaddleOCR
In [ ]!git clone https://gitee.com/PaddlePaddle/PaddleOCR -b release/2.1 --depth 1登录后复制
安装依赖环境
In [ ]!pip install imgaug pyclipper lmdb Levenshtein登录后复制
下载预训练模型
In [ ]%cd ~/PaddleOCR!wget -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_rec_pre.tar!cd pretrain_models && tar -xf ch_ppocr_mobile_v2.0_rec_pre.tar && rm -rf ch_ppocr_mobile_v2.0_rec_pre.tar登录后复制
模型训练
In [8]%cd ~/PaddleOCR!python tools/train.py -c ../multi_mnist.yml登录后复制
模型导出
In [34]%cd ~/PaddleOCR!python3 tools/export_model.py \ -c ../multi_mnist.yml -o Global.pretrained_model=../output/multi_mnist/best_accuracy \ Global.load_static_weights=False \ Global.save_inference_dir=../inference/multi_mnist登录后复制
采样测试图片
In [45]%cd ~/PaddleOCR!mkdir ~/test_imgsimport cv2import randomimport numpy as npfrom tqdm import tqdmfrom paddle.vision.datasets import MNIST# 加载数据集mnist_test = MNIST(mode='test', backend='cv2')# 数据集预处理datas_test = {}for i in range(len(mnist_test)): sample = mnist_test[i] x, y = sample[0], sample[1] _sum = np.sum(x, axis=0) _where = np.where(_sum > 0) x = 255 - x[:, _where[0][0]: _where[0][-1]+1] if str(y[0]) in datas_test: datas_test[str(y[0])].append(x) else: datas_test[str(y[0])] = [x]# 图片拼接采样for num in range(0, 1000): imgs = [255 - np.zeros((28, np.random.randint(10)))] for word in str(num): index = np.random.randint(0, len(datas_test[word])) imgs.append(datas_test[word][index]) imgs.append(255 - np.zeros((28, np.random.randint(10)))) img = np.concatenate(imgs, 1) cv2.imwrite('../test_imgs/%03d.webp' % num , img)登录后复制 模型测试
In [46]%cd ~/PaddleOCR!python tools/infer/predict_rec.py \ --image_dir="../test_imgs" \ --rec_model_dir="../inference/multi_mnist/" \ --rec_image_shape="3, 28, 64" \ --rec_char_type="ch" \ --rec_char_dict_path="../label_list.txt"登录后复制