View Code? Open in Web Editor
NEW
采样FCRN: Fully-Convolutional Regression Network (全卷积回归网络),出自VGG 实验室这篇 CVPR2016的Paper:Synthetic Data for Text Localisation in Natural Images的文本检测方法,该方法没有复现成功,但是里面涉及的斜的boundingbox可以参考,期待作者公开源码。
Shell 0.42%
Python 99.58%
fcrn_detect's People
Contributors
Watchers
fcrn_detect's Issues
hi @zj463261929 ,
感谢你的分享,不知道你是否尝试使用自己的图像制作数据集,不是采用作者提供的41G训练数据集。
edgar
Epoch 118/118
249/249 [==============================] - 19s - loss: 7.6904e-05 - acc: 0.2372 - val_loss: 1.6073e-04 - val_acc: 0.2700
Epoch 119/119
249/249 [==============================] - 19s - loss: 7.9958e-05 - acc: 0.2334 - val_loss: 2.2139e-04 - val_acc: 0.2357
Epoch 120/120
249/249 [==============================] - 19s - loss: 7.6645e-05 - acc: 0.2321 - val_loss: 2.3720e-04 - val_acc: 0.2396
Epoch 121/121
249/249 [==============================] - 19s - loss: 7.2970e-05 - acc: 0.2364 - val_loss: 4.3272e-04 - val_acc: 0.2266
Epoch 122/122
249/249 [==============================] - 19s - loss: 7.3489e-05 - acc: 0.2378 - val_loss: 2.7900e-04 - val_acc: 0.2469
Epoch 123/123
249/249 [==============================] - 19s - loss: 6.7133e-05 - acc: 0.2376 - val_loss: 3.6181e-04 - val_acc: 0.2227