search

ZJULearning / pixel_link

Implementation of our paper 'PixelLink: Detecting Scene Text via Instance Segmentation' in AAAI2018
MIT License
768 stars 255 forks source link

Can't reach the results on the paper #47

Closed ccnankai closed 6 years ago

ccnankai commented 6 years ago

Hi, @DelightRun I used my own data set for training, the test results are not good. I have trained 20,000 steps, and The loss function stops and does not reduced( stop at 0.6). Can you give some Suggestions for improvement to get better results？Which parameters need to be changed？

image

Thank you. @comzyh

ghost commented 6 years ago

Hi, How do you train that ~~I got a problem:

  1. 刚开始时loss就很低0.0010左右，等到2000step时就成0.0000了
  2. 训练的结果，只是框了红线部分，其他没有框 image 多谢您了~~
ccnankai commented 6 years ago

@greatgeekgrace 不好意思，我现在也遇到了同样问题，框的不准确。。。。。

ghost commented 6 years ago

@ccnankai 您训练的参数相对于原始的改动 辛苦告诉我一下呢~~我现在一直是没有结果呢

ccnankai commented 6 years ago

@greatgeekgrace 我记得应该不用调什么，把样本标记好就可以了。

MarStarck commented 6 years ago

@greatgeekgrace 我训练到18 step的时候loss就是0了，蛋疼

ccnankai commented 6 years ago

@MarStarck 我训练损失函数都降不下来

Simon-LLong commented 6 years ago

@MarStarck 我训练损失函数都降不下来

很难下降不知道为啥

DelightRun commented 5 years ago

@ccnankai @MarStarck @EvenIsLee Hi, all. Sorry for the late reply. Actually I and @comzyh are both not the original author of this project, please contact @dengdan.