OpenDriveLab / OpenLane

[ECCV 2022 Oral] OpenLane: Large-scale Realistic 3D Lane Dataset
Apache License 2.0
494 stars 46 forks source link

Unaccuracy uv coordinate labels #72

Open hanweijing opened 9 months ago

hanweijing commented 9 months ago

Hi, thx for excellent datasets! But when I use uv labels in json files, I draw them in image below. Why the points are so disorganized? In your paper, you mentioned about 2d/3d label generation workflow. And the first step told us the high accuracy 2d lane label is neccessary. Do I misunderstand? Hope get some explanations. 无标题

RicardLee commented 9 months ago

Hi, considering that OpenLane is focused on the task of 3D lane detection, the 2D lane line labels released to the public are not obtained by direct labelling, but by a comprehensive consideration of the two results of 3D lane lines and 2D lane lines to ensure the accuracy of its three-dimensional labels. Therefore, the 2D lane lines is different from other lane marking dataset. best, Yang Li


From: hanweijing @.> Sent: Tuesday, January 9, 2024 5:25 PM To: OpenDriveLab/OpenLane @.> Cc: Subscribed @.***> Subject: [OpenDriveLab/OpenLane] Unaccuracy uv coordinate labels (Issue #72)

Hi, thx for excellent datasets! But when I use uv labels in json files, I draw them in image below. Why the points are so disorganized? In your paper, you mentioned about 2d/3d label generation workflow. And the first step told us the high accuracy 2d lane label is neccessary. Do I misunderstand? Hope get some explanations. default.jpg (view on web)https://github.com/OpenDriveLab/OpenLane/assets/13847565/709be858-3ebe-4328-997b-0456598830bc

― Reply to this email directly, view it on GitHubhttps://github.com/OpenDriveLab/OpenLane/issues/72, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AUA46BSQ3IDWHB6UHT7DF6LYNUEJRAVCNFSM6AAAAABBSYYKTSVHI2DSMVQWIX3LMV43ASLTON2WKOZSGA3TCOJWGEZTKNY. You are receiving this because you are subscribed to this thread.Message ID: @.***>