Skip to content

Latest commit

 

History

History
 
 

lstr

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 

LSTR: Lane Shape Prediction with Transformers

Input

Images with the same aspect ratio as 360×640. This model detect lane from inputs.

Output

Images which the detected lane in input images is colored in green.

  • output image
    Four green lines mean each detected lane which model predicted and numbers attached to each line mean line number.
    Predicted curve parameters are showed at the top of the image. 出力画像

Usage

You have to specify input filetype and input filepath.

  • Image mode (image to image)
    You run sample script as below if your desired file path is {Path to LSTR}/input/image/1.jpg.
$ python3 lstr.py --input input.jpg
  • Video mode (video to video)
    You run sample script as below if your desired file path is {Path to LSTR}/input/video/video1.mp4.
$ python3 lstr.py -v input.mp4

Reference

  • Repository
    LSTR

  • Input images and videos
    Input images are part of TuSimple dataset and input videos are created by using TuSimple images.

Framework

PyTorch 1.8.0

Model Format

ONNX opset = 11

Netron

lstr.onnx.prototxt