pan_pp.pytorch

pan_pp.pytorch copied to clipboard

Hi, thanks for your awesome work. Could you please publish the pretrain model of pan++?

mxin262

您好，我注意到在pan的训练过程中，通过 SynthText 预训练模型，最终效果有了一定的提升。 我想请问在config文件中 SynthText 的配置程序的 train_cfg 中的 epoch 被设置为了 1 ，请问这是正常情况吗？因为在后续 finetune 过程中的 epoch 均被设置为 600 。我在考虑二者差距是否太过悬殊？ 期待您的解答，谢谢！

mashumli

UnicodeDecodeError: 'utf-8' codec can't decode byte 0xb5 in position 35: invalid start byte 用中文的自己制作的数据集训练ic15报错，有解决办法吗

qqqqq127

Too many bboxes detection during testing

6

Thanks for your work. when I tried to test PAN++ with poly type network produce too many polygon points. which is not similar to total text ground truth file. could...

bharatsubedi

the FAST is amazing,have you tested the FPS when not using the GPU-parallel post-processing(change to CPU) in FAST?

1

xiongzl

PA function is much slower than pse function during my imgs

1

test on a test dataset of small imgs (about 600 pics)，panet takes more time than psenet, average 20ms higher than psenet test on a test dataset of big imgs (about...

Zhang-O

When release the "fast" code?

3

Hi, Your FAST paper is really amazing! I wish try it on my own tasks, when release the "fast" code? Looking forward your reply! Thanks!

zhangyuan1994511

https://github.com/whai362/pan_pp.pytorch/blob/aa4774b1bf360d0a8e54d520483514d57521bf34/models/pan.py#L62 Through my test in doc scene, FFM of pan is better than FFM of pan_pp when using the same model trained in a way of pan_pp. Why you changed...

Zhang-O

can not download the pretained of resnet

7

感谢作者对代码的开源，目前我遇到一个问题：Resnet预训练文件通过该网址无法加载，连接超时？ 在colab和本地(尝试过开启或关闭VPN)都无法成功加载 

kongbailongtian

关于中文识别

1

请问Pan++支持中文的识别吗？自己测试对于中文都无法识别是什么原因，谢谢!

Floyd809