Test on a single input image

[hthanhle]

Anyone can show me how to run the code to test on a single input image? Any comments and suggestions would be much appreciated.

[farhuaa]

https://github.com/Turoad/lanedet/blob/main/tools/detect.py

[hthanhle]

Thanks so much. It's really helpful.

[hthanhle]

It seems that the model works well with images from the dataset used for training (see the following screenshot).

However, it does not work with real images (see the following screenshot).

Could anyone give me any suggestions?

[mengxia1994]

I met the same problem. I think it is the image size problem. I resize my images to 1280*720 like tusimple and I get the right output. If you don't resize, it won't come out an error but the result of detect.py is not right.

[hthanhle]

Thanks for your comments, @mengxia1994.

I already resized the input image to 1280 x 720, but it does not work (see the 2nd image in my previous post). Following is my configuration:

• Config: 'configs/clrnet/clr_resnet18_tusimple.py'
• Weights: 'tusimple_r18.pth'

Could you please tell me how you can resolve the problem?

Also, I found that the parameter 'cut_height' is really tricky, that is only applicable for the existing datasets. What is the appropriate value if we test on a new image?

[jinhoyoho]

Custom data was available for CLRNet by setting 'cut_height = 0' in configs/clrnet/

[mengxia1994]

您好，我已收到您的邮件，我将尽快查阅并给予回复，谢谢！（孟夏的自动回复）