CULane evaluation problem

harryhan618 / SCNN_Pytorch

Pytorch implementation of "Spatial As Deep: Spatial CNN for Traffic Scene Understanding"

MIT License

250 stars 68 forks source link

CULane evaluation problem #64

Closed yezi-yang closed 3 years ago

yezi-yang commented 3 years ago

Hi,

I'm using the lane evaluation code in utils to measure the f1 score, tp, fp and fn. But I realise that the total number of lines is smaller that the one in the ground truth. For example, if I have 1000 lines in the ground truth, but the tp+fn is only 617. Does this also happen to someone else?