def __init__(self, cfg_file='/root/tzl/text_renderer/configs/default_test.yaml'):
"""
handle exist image
"""
self.debug = True
cfg = load_config(cfg_file)
self.cfg = cfg
self.liner = Liner(cfg)
self.noiser = Noiser(cfg)
self.remaper = Remaper(cfg)
self.create_kernals()
def __init__(self, cfg_name):
self.cfg = load_config(cfg_name)
self.cfg.lr_boundaries = [10000]
self.cfg.lr_values = [
self.cfg.lr * (self.cfg.lr_decay_rate**i)
for i in range(len(self.cfg.lr_boundaries) + 1)
]
self.sess = None
self.graph = None
self.converter = None
self.dataset = None
self.input = None
self.output = None
self.global_step = None
def __init__(self, model_path):
os.environ['CUDA_VISIBLE_DEVICES'] = '2'
self.cfg = load_config('resnet')
self.label_converter = LabelConverter(lexicon)
config = tf.ConfigProto(allow_soft_placement=True)
config.gpu_options.allow_growth = True
config.gpu_options.per_process_gpu_memory_fraction = 0.5
self.graph = tf.Graph()
self.session = tf.Session(graph=self.graph, config=config)
with self.session.as_default():
with self.graph.as_default():
self.net = CRNN(self.cfg, num_classes=self.label_converter.num_classes)
saver = tf.train.Saver()
saver.restore(self.session, model_path)
logging.info('CRNN model initialized.')
def __init__(self, args):
self.args = args
self.cfg = load_config(args.cfg_name)
self.converter = LabelConverter(chars_file=args.chars_file)
self.tr_ds = ImgDataset(args.train_dir, self.converter,
self.cfg.batch_size)
self.cfg.lr_boundaries = [
self.tr_ds.num_batches * epoch
for epoch in self.cfg.lr_decay_epochs
]
self.cfg.lr_values = [
self.cfg.lr * (self.cfg.lr_decay_rate**i)
for i in range(len(self.cfg.lr_boundaries) + 1)
]
if args.val_dir is None:
self.val_ds = None
else:
self.val_ds = ImgDataset(args.val_dir,
self.converter,
self.cfg.batch_size,
shuffle=False)
if args.test_dir is None:
self.test_ds = None
else:
# Test images often have different size, so set batch_size to 1
self.test_ds = ImgDataset(args.test_dir,
self.converter,
shuffle=False,
batch_size=1)
self.model = CRNN(self.cfg, num_classes=self.converter.num_classes)
self.sess = tf.Session(config=tf.ConfigProto(
allow_soft_placement=True))
self.epoch_start_index = 0
self.batch_start_index = 0
from libs.config import load_config
from libs.timer import Timer
from parse_args import parse_args
import libs.utils as utils
import libs.font_utils as font_utils
from textrenderer.corpus.corpus_utils import corpus_factory
from textrenderer.renderer import Renderer
from tenacity import retry
lock = mp.Lock()
counter = mp.Value('i', 0)
STOP_TOKEN = 'kill'
flags = parse_args()
cfg = load_config(flags.config_file)
fonts = font_utils.get_font_paths_from_list(flags.fonts_list)
bgs = utils.load_bgs(flags.bg_dir)
corpus = corpus_factory(flags.corpus_mode, flags.chars_file, flags.corpus_dir,
flags.length)
renderer = Renderer(corpus,
fonts,
bgs,
cfg,
height=flags.img_height,
width=flags.img_width,
clip_max_chars=flags.clip_max_chars,
debug=flags.debug,
def train(self, log_dir, restore, log_step, ckpt_dir, val_step, cfg_name, chars_file, train_txt, val_txt, test_txt, result_dir):
cfg = load_config(cfg_name)
converter = LabelConverter(chars_file=chars_file)
tr_ds = ImgDataset(train_txt, converter, cfg.batch_size)
cfg.lr_boundaries = [10000]
cfg.lr_values = [cfg.lr * (cfg.lr_decay_rate ** i) for i in
range(len(cfg.lr_boundaries) + 1)]
if val_txt is None:
val_ds = None
else:
val_ds = ImgDataset(val_txt, converter, cfg.batch_size, shuffle=False)
if test_txt is None:
test_ds = None
else:
# Test images often have different size, so set batch_size to 1
test_ds = ImgDataset(test_txt, converter, shuffle=False, batch_size=1)
model = CRNN(cfg, num_classes=converter.num_classes)
epoch_start_index = 0
batch_start_index = 0
config=tf.ConfigProto(allow_soft_placement=True)
config.gpu_options.allow_growth = True
config.gpu_options.per_process_gpu_memory_fraction = 0.8
sess = tf.Session(config=config)
sess.run(tf.global_variables_initializer())
saver = tf.train.Saver(tf.global_variables(), max_to_keep=8)
train_writer = tf.summary.FileWriter(log_dir, sess.graph)
if restore:
self._restore(sess, saver, model,tr_ds, ckpt_dir)
print('Begin training...')
for epoch in range(epoch_start_index, cfg.epochs):
sess.run(tr_ds.init_op)
for batch in range(batch_start_index, tr_ds.num_batches):
batch_start_time = time.time()
if batch != 0 and (batch % log_step == 0):
batch_cost, global_step, lr = self._train_with_summary( model, tr_ds, sess, train_writer, converter)
else:
batch_cost, global_step, lr = self._train(model, tr_ds, sess)
print("epoch: {}, batch: {}/{}, step: {}, time: {:.02f}s, loss: {:.05}, lr: {:.05}"
.format(epoch, batch, tr_ds.num_batches, global_step, time.time() - batch_start_time,
batch_cost, lr))
if global_step != 0 and (global_step % val_step == 0):
val_acc = self._do_val(val_ds, epoch, global_step, "val", sess, model, converter, train_writer, cfg, result_dir)
test_acc = self._do_val(test_ds, epoch, global_step, "test", sess, model, converter, train_writer, cfg, result_dir)
self._save_checkpoint(ckpt_dir, global_step, saver, sess, val_acc, test_acc)
batch_start_index = 0
self.embedding: embedding features
self.char_label: 和 embedding features 对应的标签
self.char_pos: 和 embedding features 对应的字符位置
"""
with tf.variable_scope('pos'):
# 判断是否为预测的字符
is_char = tf.less(raw_pred, self.num_classes - 1)
# 错位比较法，找到重复字符
char_rep = tf.equal(raw_pred[:, :-1], raw_pred[:, 1:])
tail = tf.greater(raw_pred[:, :1], self.num_classes - 1)
char_rep = tf.concat([char_rep, tail], axis=1)
# 去掉重复字符之后的字符位置，重复字符取其 最后一次 出现的位置
char_no_rep = tf.math.logical_and(is_char,
tf.math.logical_not(char_rep))
# 得到字符位置 和 相应的标签，如果某张图片 预测出来的字符数量 和gt不一致则跳过
self.char_pos, self.char_label = self.get_char_pos_and_label(
preds=char_no_rep, label=label, char_num=char_num, poses=poses)
# 根据字符位置得到字符的 embedding
self.embedding = self.get_features(self.char_pos, embedding)
if __name__ == '__main__':
from libs.label_converter import LabelConverter
cfg = load_config('raw')
converter = LabelConverter(chars_file='./data/chars/lexicon.txt')
model = CRNN(cfg, num_classes=converter.num_classes)
async def on_ready(self):
print(f'We have logged in as {self.bot.user}')
await self.bot.change_presence(activity=discord.Game(name=config.special_event))
config.load_config(self.bot)
print("Session started.")
