XingangPan / SCNN

Spatial CNN for traffic lane detection (AAAI2018)
MIT License
815 stars 242 forks source link

DL vs CV #113

Open Mouiad-JRA opened 4 years ago

Mouiad-JRA commented 4 years ago

Hi there, my project is a comparison between computer vision and deep learning for self-driving cars and i'm using your model for DL. The problem I'm currently facing is when I compare the two parts (DL vs CV) together through IoU and F-measure , the datasets labels (imagename.line.txt) which contains the lines coordinates (the end points) are usually three or Four lines of points in the line.txt files, this labels are very good for the SCNN but for the CV its very bad, to improve that, the labels must be 2 lines of points not four for each image. How Can i do that? and if that possible , it will effect badly for the SCNN?