def get_frame():
tic = tc()
print('----------------------------------------------------------------')
log_time = time.strftime('%Y-%m-%d %H:%M:%S')
print(log_time)
cap_file = request.files['image']
cap_img = Image.open(cap_file)
word_ind, objs_ind = model.predict(cap_img)
Result = {'Result': []}
for i, ind in enumerate(objs_ind):
if ind == word_ind[0]:
Result['Result'].append(i + 1)
if Result['Result'].__len__() <= 0:
Result['Result'].append(np.random.randint(1, 9))
print(classes[word_ind[0]], Result['Result'])
pic = log_time.replace(' ', '').replace('-', '').replace(
':', '') + '_' + classes[word_ind[0]] + '_' + '_'.join(
[str(v) for v in Result['Result']]) + '.png'
print(pic)
cap_img.save(os.path.join(logdir, pic))
print('fetch time: {:.5f} s.'.format(tc() - tic))
print('----------------------------------------------------------------\n')
return json.dumps(Result)
def crack_captcha(captcha_img):
tic = tc()
print("模型加载完毕")
word_ind, objs_ind = model.predict(captcha_img)
word = classes[word_ind]
objs = classes[objs_ind]
print(unicode(word[0], "utf-8"), objs[0])
print('time elapsed:{:.2f}'.format(tc()-tic))
return word[0]
def file_merge(src_path, dst_path, batch_size=100, is_training=True):
intensity_path = os.path.join(src_path, 'intensity')
pts_path = os.path.join(src_path, 'pts')
category_path = os.path.join(src_path, 'category')
fns = [fn for fn in os.listdir(pts_path) if fn[-4:] == '.csv']
n_files = len(fns)
flame_index = 0
tic = tc()
for batch_i in range(int(np.ceil(n_files / batch_size))):
print('batch: {}/{} {:.2f}s.'.format(batch_i + 1, round(n_files / batch_size + 0.5), tc() - tic))
start = batch_i * batch_size
end = start + batch_size
index, pts, intensity, category = [], [], [], []
test_fn_index = []
for fn in fns[start:end]:
tmp1 = np.loadtxt(os.path.join(pts_path, fn), dtype=np.float32, delimiter=',')
pts.append(tmp1)
tmp2 = np.loadtxt(os.path.join(intensity_path, fn), dtype=np.float32).reshape((-1, 1))
intensity.append(tmp2)
if is_training:
category.append(np.loadtxt(os.path.join(category_path, fn), dtype=np.float32).reshape((-1, 1)))
index.append(flame_index*np.ones_like(tmp2, np.float32))
test_fn_index.append([flame_index, fn])
flame_index += 1
index = np.vstack(index)
pts = np.vstack(pts)
intensity = np.vstack(intensity)
if is_training:
category = np.vstack(category)
merged_data = np.hstack((index, pts, intensity, category))
save_path = os.path.join(dst_path, 'train_batch_{}.npy'.format(str(batch_i)))
else:
merged_data = np.hstack((index, pts, intensity))
save_path = os.path.join(dst_path, 'test_batch_{}.npy'.format(str(batch_i)))
pd.DataFrame(test_fn_index).to_csv(
os.path.join(dst_path, 'test_batch_{}.name'.format(str(batch_i))),
header=None,
index=False,
encoding='utf-8'
)
np.float32(merged_data).tofile(save_path)
with tf.Session() as sess:
saver = tf.train.Saver(tf.global_variables(), max_to_keep=20)
sess.run(init)
# tensorboard 可视化
merged = tf.summary.merge_all()
summary_writer_train = tf.summary.FileWriter('logs/train', sess.graph)
summary_writer_test = tf.summary.FileWriter('logs/test')
for epoch in range(1, total_epochs + 1):
if epoch in reduce_lr_epoch:
epoch_learning_rate /= 10
# 训练
tic = tc()
train_acc, train_loss = train_epoch(sess, epoch_learning_rate)
train_time = tc() - tic
# 测试
test_acc, test_loss = eva(sess)
print('\nepoch: {}, training time: {:.0f} s.'.format(
epoch, train_time))
print('train loss: {:.5f}, accuracy: {:.3f}'.format(
train_loss, train_acc))
print('test loss: {:.5f}, accuracy: {:.3f}'.format(
test_loss, test_acc))
history.append(
[epoch, train_acc, train_loss, test_acc, test_loss,
# 读取数据块
chunk_data, chunk_type_code = png_read_chunk(fp)
if chunk_type_code not in chunk_type.keys():
chunk_type[chunk_type_code] = 1
else:
chunk_type[chunk_type_code] += 1
print(image_info)
print(chunk_type)
if __name__ == "__main__":
tic = tc()
block_size = (512, 512)
# ------------- train data ----------------
# train image_1.png
png_file = '../data/jingwei_round1_train_20190619/image_1.png'
# check_png(png_file)
png_sub(png_file, '../data/sub/train_1', block_size)
print('train image_1.png', tc() - tic)
# train image_2.png
png_file = '../data/jingwei_round1_train_20190619/image_2.png'
# check_png(png_file)
png_sub(png_file, '../data/sub/train_2', block_size)
print('train image_2.png', tc() - tic)
ocr_model_meta = './model_data/ocr.ckpt-1108000.meta'
ocr_model_para = './model_data/ocr.ckpt-1108000'
cap_model_meta = './model_data/densenet_121.ckpt-145000.meta'
cap_model_para = './model_data/densenet_121.ckpt-145000'
classes_path = './model_data/12306_classes.txt'
classes = read_classes(classes_path)
model = CaptchaModel(ocr_model_meta=ocr_model_meta,
ocr_model_para=ocr_model_para,
cap_model_meta=cap_model_meta,
cap_model_para=cap_model_para)
from time import time as tc
tic = tc()
print("load model done")
for x in range(2, 19):
word_ind, objs_ind = model.predict(
'/Users/dtstack/PycharmProjects/Crack12306WithAI/misc/captcha/e_{}.jpg'
.format(str(x)))
word = classes[word_ind]
objs = classes[objs_ind]
print("e_{}.jpg".format(str(x)), unicode(word[0], "utf-8"), objs[0])
print('time elapsed:{:.2f}'.format(tc() - tic))
model.close()
print("Is there a prime between 1328 and 1361:",
any(is_prime(x) for x in range(1328, 1361)))
monday = [11, 12, 13, 14, 15, 16, 17, 17, 17, 16, 16, 15, 14, 13, 12, 11, 11]
tuesday = [x * 2 - 10 for x in monday]
print(monday, tuesday)
for item in zip(monday, tuesday):
print(item, type(item))
for d1, d2 in zip(monday, tuesday):
print(f"Hourly average is {(d1 + d2)/2}°C")
wednesday = [x * 2 - 20 for x in tuesday]
for temps in zip(monday, tuesday, wednesday):
print(
f"min={min(temps):4.1f}\t max={max(temps):4.1f}\t avg={sum(temps)/len(temps):4.1f}"
)
from itertools import chain
temperatures = chain(monday, tuesday, wednesday)
print(monday, tuesday, wednesday) # concatenation
print(list(temperatures)) # lazy concatenation
from md_lucas import lucas
from time import perf_counter as tc
start = tc()
for x in (p for p in lucas() if is_prime(p)):
print(x, "time:", tc() - start)
sess.run(init)
if ckpt and tf.train.checkpoint_exists(ckpt.model_checkpoint_path):
print('loading model {}'.format(ckpt.model_checkpoint_path))
saver.restore(sess, ckpt.model_checkpoint_path)
global_step = int(ckpt.model_checkpoint_path.split('-')[-1]) + 1
else:
global_step = 0
if is_training:
merged = tf.summary.merge_all()
summary_writer_train = tf.summary.FileWriter(log_dir + '/train',
sess.graph)
summary_writer_valid = tf.summary.FileWriter(log_dir + '/valid')
tic = tc()
epoch_learning_rate = init_learning_rate
for epoch in range(total_epochs):
local_step = 0
train_idxs = np.random.permutation(499)
for i in range(499):
print('flame: {}/{} '.format(i + 1, 499))
data, _ = npy_item_read(train_path, train_idxs[i])
flame_idxs = np.unique(data[:, 0])
np.random.shuffle(flame_idxs)
for flame in flame_idxs:
# 处理每一帧数据
lg = data[:, 0] == flame
pts, intensity, category = data[lg, 1:4], data[lg,