search

hirotomusiker / CLRerNet

The official implementation of "CLRerNet: Improving Confidence of Lane Detection with LaneIoU"
Apache License 2.0
181 stars 19 forks source link

redundant #49

Closed 10qiu01 closed 1 month ago

10qiu01 commented 1 month ago

  1. I tried to process a screenshot of my video data, but the resulting image is very weird. I was wondering if there is any restrictions for input? By the way, did you use padding during resize or cropping the input image?

  2. Do you have a mature pipeline for video data? Thanks in advance!!!

hirotomusiker commented 1 month ago

1: Could you refer to the following issue for cropping settings? https://github.com/hirotomusiker/CLRerNet/issues/34

(1) Modify the crop boxso that the coordinates [x0, y0, x1, y1] are within your input shape and the vanishing point is within the cropped window. (2) Modify test_cfg accordingly.

For example, if you wish to use (1920, 1080) format without cropping, it'd be like this:

crop_bbox = [0, 0, 1920, 1080]
test_cfg=dict(
...
    nms_topk=4,
    ori_img_w=1920,
    ori_img_h=1080,
    cut_height=0,
)
  1. Please refer to the following issue. https://github.com/hirotomusiker/CLRerNet/issues/33