cfzd / Ultra-Fast-Lane-Detection

Ultra Fast Structure-aware Deep Lane Detection (ECCV 2020)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

number of grid

gebawe opened this issue · comments

commented

Thank you for the great job.
I have noticed that the number of gird sample are 200 in the config parameter, however the model output has griding_num=201 (i.e., Culane dataset: the shape is 18x201x4). Can somebody explain why they have added +1 (cfg.griding_num+1) into the griding_num when initialling the network?

net = parsingNet(pretrained = False, backbone=cfg.backbone,cls_dim = (cfg.griding_num+1, cls_num_per_lane, 4),
                    use_aux=False).cuda() 
commented

@awethaileslassie
The 200 grids are used for localization. The extra one grid is used for represneting existence of lane points.