def main():
# 读取训练用数据
print('Start: read data', time.process_time())
X = pretreatment.load_data() # NOQA
c, h, w = X.shape
X = X.reshape((c, h * w)) # NOQA
X = X / 255.0 # NOQA
print('Samples', len(X), 'Feature', len(X[0]))
# PCA
print('Start: PCA', time.process_time())
pca = PCA(n_components=0.99)
pca.fit(X)
X = pca.transform(X) # NOQA
print('Samples', len(X), 'Feature', len(X[0]))
sys.stdout.flush()
# 训练
print('Start: train', time.process_time())
n_clusters = 2000 # 聚类中心个数
estimator = KMeans(n_clusters, n_init=1, max_iter=20, verbose=True)
estimator.fit(X)
print('Clusters', estimator.n_clusters, 'Iter', estimator.n_iter_)
print('Start: classify', time.process_time())
np.save('labels.npy', estimator.labels_)
print('Start: save model', time.process_time())
joblib.dump(estimator, 'k-means.pkl')
def main():
import sys
from pretreatment import load_data
texts = load_data()
fp = open('texts.log', 'w', encoding='utf-8')
for idx, text in enumerate(texts):
try:
text = ocr(text)
print(idx, text, file=fp)
print(idx, text)
except Exception as e:
print(e, file=sys.stderr)
def learn():
texts, imgs = load_data()
labels = mlearn.predict(texts)
labels = labels.argmax(axis=1)
imgs.dtype = np.uint64
imgs.shape = (-1, 8)
unique_imgs = np.unique(imgs)
print(unique_imgs.shape)
imgs_labels = []
for img in unique_imgs:
idxs = np.where(imgs == img)[0]
counts = np.bincount(labels[idxs], minlength=80)
imgs_labels.append(counts)
np.savez('images.npz', images=unique_imgs, labels=imgs_labels)
def main():
import sys
from pretreatment import load_data
texts, _ = load_data()
fp = open('texts.log', 'w', encoding='utf-8')
for idx, text in enumerate(texts):
try:
# cv2.imshow('lena', text)
# cv2.waitKey(1000)
text = ocr(text)
print(idx, text, file=fp)
print(idx, text)
except Exception as e:
print(e, file=sys.stderr)
# coding: utf-8
import sys
import cv2
import numpy as np
import pretreatment
import utils
result_fn = sys.argv[1]
classify_fn = sys.argv[2]
utils.mkdir(classify_fn)
# 用于统计有多少聚类中心是有样本的
result = np.load(result_fn)
print(np.unique(result).shape)
# 将聚类后的样本复制并使用聚类结果命名
imgs = pretreatment.load_data()
for idx, (img, classify) in enumerate(zip(imgs, result)):
dst = f'{classify_fn}/{classify}({idx}).jpg'
cv2.imwrite(dst, img)