Automated lane boundary annotations for the Malaysian urban and highway driving dataset

This dataset provides automated, per frame lane boundary annotations for the 27 video clips released in the companion Malaysian Urban and Highway Driving Dataset by the same author on Mendeley Data: Clip #1: https://doi.org/10.17632/rkyfwv9hw8.1 Clip #2: https://doi.org/10.17632/ynmympptmy.1 Clip #3: https://doi.org/10.17632/m88955h7vb.1 Clip #4: https://doi.org/10.17632/vry8n5nkrf.1 Clip #5: https://doi.org/10.17632/f24x2p6b5h.3 Clip #6: https://doi.org/10.17632/xskxs82mz6.3 Clip #7: https://doi.org/10.17632/dppstzh8n6.4 Clip #8: https://doi.org/10.17632/hgt5whhj6n.3 Clip #9: https://doi.org/10.17632/bvbykc4hxf.4 Clip #10: https://doi.org/10.17632/g98zzcn6nr.3 Clip #11: https://doi.org/10.17632/z3yjbd4567.3 Clip #12: https://doi.org/10.17632/ytn823rw8j.3 Clip #13: https://doi.org/10.17632/946jzttn7n.3 Clip #14: https://doi.org/10.17632/cww75348bj.3 Clip #15: https://doi.org/10.17632/k74tdgbhjm.3 Clip #16: https://doi.org/10.17632/hps9jsjwxp.4 Clip #17: https://doi.org/10.17632/bxmmttx535.3 Clip #18: https://doi.org/10.17632/smx7tbx29p.3 Clip #19: https://doi.org/10.17632/kcxpm835gw.3 Clip #20: https://doi.org/10.17632/m25z57438h.3 Clip #21: https://doi.org/10.17632/cjptbmddpk.4 Clip #22: https://doi.org/10.17632/yhd2j7ddxc.3 Clip #23: https://doi.org/10.17632/5zjf62drv7.3 Clip #24: https://doi.org/10.17632/r8vm7nbgvm.3 Clip #25: https://doi.org/10.17632/642n3xx8s6.3 Clip #26: https://doi.org/10.17632/wmymrk79tg.3 Clip #27: https://doi.org/10.17632/wb4hgnr6k3.3 Annotations were generated using YOLOP, a pretrained deep learning model for lane and drivable area segmentation, run at 5 frames per second across all 27 clips, giving 11,665 annotated frames in total. For each sampled frame, the release provides: pixel coordinates of the top and bottom points of the left and right ego lane boundary; a lateral offset ratio and derived lane departure flag, computed using the same definition as the vision based lane departure warning framework validated on this dataset in Poh Ping et al. (2019, Heliyon, https://doi.org/10.1016/j.heliyon.2019.e02169); a detection_status field recording whether both, one, or neither lane boundary was found in that frame; the same annotations in TuSimple format for direct use in common lane detection training pipelines; the corresponding extracted JPEG frame for every annotated sample; a set of quality assurance overlay images for visual auditing; and a metadata file documenting the full method, licensing, sampling scheme, and field definitions. Across the 11,665 annotated frames, both lane boundaries were detected in 99.8 percent of frames. These are silver labels: they are generated automatically and have not been manually verified, and should be treated as a starting point for training or benchmarking rather than as ground truth. Full detection quality statistics and known limitations are documented in the included metadata file. Released under CC BY 4.0, consistent with the rest of the dataset.