harryhan618 / SCNN_Pytorch

Pytorch implementation of "Spatial As Deep: Spatial CNN for Traffic Scene Understanding"
MIT License
250 stars 68 forks source link

About the experiments results #3

Closed lHou157 closed 5 years ago

lHou157 commented 5 years ago

Hello,thank you for your sharing! Could you please add some supplying about the experiments results such as accuracy or F1-measure and so on if possible,I have no idea about it and I have been stuck on this for many days. I really appreciate it if you could share your thought or ideas. Thank you again!

harryhan618 commented 5 years ago

Sure. I would add evalution codes and descripion later these days. BTW, there's a revision in the network. Stay updated.

lHou157 commented 5 years ago

Oh! Thank you very much! I am looking forward to your sharing! You are really a nice man!

harryhan618 commented 5 years ago

There's an important change in the model.py. Please get updated.

lHou157 commented 5 years ago

Oh! Thank you very much and I will study it carefully!

发送自 Windows 10 版邮件应用

发件人: HarryHan 发送时间: 2019年5月9日 11:45 收件人: harryhan618/SCNN_Pytorch 抄送: lHou157; Author 主题: Re: [harryhan618/SCNN_Pytorch] About the experiments results (#3)

There's an important change in the model.py. Please get updated. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or mute the thread.

ZhiyuanChen commented 5 years ago

Sure. I would add evalution codes and descripion later these days. BTW, there's a revision in the network. Stay updated.

Hi, thank you very much for your work. I tried some images from TuSimple dataset and my own test images, however, it seems the nn cannot detect the lanes at all. Is there any issue with pre-trained model?

harryhan618 commented 5 years ago

Sure. I would add evalution codes and descripion later these days. BTW, there's a revision in the network. Stay updated.

Hi, thank you very much for your work. I tried some images from TuSimple dataset and my own test images, however, it seems the nn cannot detect the lanes at all. Is there any issue with pre-trained model?

The currently provided model is trained on CULane dataset. If you want to use it on other dataset or your images, you have to train the model on the corresponding dataset first. Tansfer issue is kind of a demerit of deep learning approach.

ZhiyuanChen commented 5 years ago

Sure. I would add evalution codes and descripion later these days. BTW, there's a revision in the network. Stay updated.

Hi, thank you very much for your work. I tried some images from TuSimple dataset and my own test images, however, it seems the nn cannot detect the lanes at all. Is there any issue with pre-trained model?

The currently provided model is trained on CULane dataset. If you want to use it on other dataset or your images, you have to train the model on the corresponding dataset first. Tansfer issue is kind of a demerit of deep learning approach.

lol, thanks for replying. I thought this as a lane line detecor, would have good generalization ability.