naurril / SUSTechPOINTS

3D Point Cloud Annotation Platform for Autonomous Driving
GNU General Public License v3.0
785 stars 206 forks source link

转换到kitti格式将3d框投影到图像上不准确 #112

Open Dorablank opened 1 year ago

Dorablank commented 1 year ago

Hi naurril

我尝试将标注好的数据转化成KITTI数据集的格式 并将3d框投影到图像上验证效果, 出现了如下问题

image

左图是转化到kitti格式投影到图像上的结果,右图是sustechpoints中将目标框投影到图像上的结果 很明显左图的角度偏了一些,下面图是转换部分代码 image

能给些建议嘛?感谢您!

naurril commented 1 year ago

不能直接这样转,需要根据外参矩阵来

Dorablank @.***> 于 2022年12月3日周六 14:34写道:

Hi naurril

我尝试将标注好的数据转化成KITTI数据集的格式 并将3d框投影到图像上验证效果, 出现了如下问题

[image: image] https://user-images.githubusercontent.com/76980848/205428152-0bb1d3f0-2612-4cdf-98ba-38710a7dddb7.png

左图是转化到kitti格式投影到图像上的结果,右图是sustechpoints中将目标框投影到图像上的结果 很明显左图的角度偏了一些,下面图是转换部分代码 [image: image] https://user-images.githubusercontent.com/76980848/205428167-c4eee841-9f97-48a8-a398-1107aae3c937.png

能给些建议嘛?感谢您!

— Reply to this email directly, view it on GitHub https://github.com/naurril/SUSTechPOINTS/issues/112, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABZKCDWTDALBYQK2MVYCLSDWLLSWVANCNFSM6AAAAAASSST35U . You are receiving this because you are subscribed to this thread.Message ID: @.***>

Dorablank commented 1 year ago

不能直接这样转,需要根据外参矩阵来 Dorablank @.> 于 2022年12月3日周六 14:34写道: Hi naurril 我尝试将标注好的数据转化成KITTI数据集的格式 并将3d框投影到图像上验证效果, 出现了如下问题 [image: image] https://user-images.githubusercontent.com/76980848/205428152-0bb1d3f0-2612-4cdf-98ba-38710a7dddb7.png 左图是转化到kitti格式投影到图像上的结果,右图是sustechpoints中将目标框投影到图像上的结果 很明显左图的角度偏了一些,下面图是转换部分代码 [image: image] https://user-images.githubusercontent.com/76980848/205428167-c4eee841-9f97-48a8-a398-1107aae3c937.png 能给些建议嘛?感谢您! — Reply to this email directly, view it on GitHub <#112>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABZKCDWTDALBYQK2MVYCLSDWLLSWVANCNFSM6AAAAAASSST35U . You are receiving this because you are subscribed to this thread.Message ID: @.>

可以再具体一些嘛?谢谢! 我使用的标定参数是data目录下面的front.json 图片是000965.jpg

naurril commented 1 year ago

tr_velo_to_cam是用的哪个, 还要看看左边的图是怎么绘制的

BaofengZan commented 1 year ago

我问一下,为什么 location中 y得值需要加 z/2?