我已經嘗試了多個車道檢測模型。使用 ResNet-18 backbone 的 CULane 模型結果波動較大,而 TuSimple 模型甚至無法顯示任何結果。在 25 幀中,有一兩幀會出現扭曲的疊加/檢測,而且結果完全不正確。非常感謝您在這方面的指導與幫助!
I have tried multiple models for lane detection. The CULane model with a ResNet-18 backbone is very unstable, and the TuSimple models do not produce any visible results. Out of 25 frames, one or two random frames show a distorted overlay/detection, which is not even correct. I would greatly appreciate your guidance on this. Thank you very much!