Open RunkaiZhao opened 7 months ago
Following the training procedure of GenLaneNet, we first train the segmentation subnetwork and then train the 3D-geometry subnetwork. No more complex strategies or tricks are used. Please provide your training details for discussion.
Great work! Thank you for your contribution to the community.
I was impressed that your re-implementation of GenLaneNet achieved a 42.8% F1 score, as mentioned in Table 2 of your paper. However, when I attempted to train the GenLaneNet using your code, the F1 score remained at zero on the validation set. Could you please provide some explanations or suggestions? Thank you in advance.