In the paper and implementation it is mentioned that the images are resized to 800x320. However, from the code it can be observed that there are additional details such as cropping. Can you write in detail you evaluation procedure for CuLane and Curvelanes? As far as I know, the culane evaluation is taken as 800x288 resolution in previous papers.
I also want to ask that how the lane pixel width is set in evaluation. What is the IoU threshold in order to assume that lane is correct for each dataset. I think these are missing in the paper and missing explanations making hard to compare with the current literature. Curvelanes have images with different resolution, is there any change in lane width for different sized images of curvelanes?
Hello, I want to ask your evaluation procedure.
In the paper and implementation it is mentioned that the images are resized to 800x320. However, from the code it can be observed that there are additional details such as cropping. Can you write in detail you evaluation procedure for CuLane and Curvelanes? As far as I know, the culane evaluation is taken as 800x288 resolution in previous papers.
I also want to ask that how the lane pixel width is set in evaluation. What is the IoU threshold in order to assume that lane is correct for each dataset. I think these are missing in the paper and missing explanations making hard to compare with the current literature. Curvelanes have images with different resolution, is there any change in lane width for different sized images of curvelanes?
Thanks in advance