hirotomusiker / CLRerNet

The official implementation of "CLRerNet: Improving Confidence of Lane Detection with LaneIoU"
Apache License 2.0
174 stars 18 forks source link

custom data #50

Open yunmengmengyun opened 1 week ago

yunmengmengyun commented 1 week ago

Dear authors,

Thanks for your work. If I want to run the models on my own images, what configurations do I need to change to achieve the best performance? currently, I know I need to change the ori img sizes and crop bbox. is there any other parameters I need to configure?

Because I found the model doesn't work well for the broken lane markings.

Thanks in advance!!!!!!!!!!!!!!

hirotomusiker commented 6 days ago

We firstly recommend to adjust (decrease) the confidence threshold to detect unclear lanes.

Second, check if the maximum number of outputs https://github.com/hirotomusiker/CLRerNet/blob/main/configs/clrernet/base_clrernet.py#L48 is more than the outputs you need to detect at once.

Thank you.