HKUST-Aerial-Robotics / Stereo-RCNN

Code for 'Stereo R-CNN based 3D Object Detection for Autonomous Driving' (CVPR 2019)
MIT License
692 stars 177 forks source link

Annotate my own 3d object dataset #56

Open Jokoe66 opened 5 years ago

Jokoe66 commented 5 years ago

Hi, thanks for your nice research. I 'd like to collect a custom dataset similar to Kitti' s. While I have a doubt that, is it true the 2d bbox we use for training is the 2d projection of its corresponding 3d bbox? If yes, then usually the 2d bboxes cover more areas than the 2d objects do when the 3d bboxes closely surround the 3d objects. In another word, the 2d bboxes are not that close to the boundaries of the 2d objects. Do I make it clear? Thank you if you could help me with this question.

lcy1996l commented 2 years ago

Hi, thanks for your nice research. I 'd like to collect a custom dataset similar to Kitti' s. While I have a doubt that, is it true the 2d bbox we use for training is the 2d projection of its corresponding 3d bbox? If yes, then usually the 2d bboxes cover more areas than the 2d objects do when the 3d bboxes closely surround the 3d objects. In another word, the 2d bboxes are not that close to the boundaries of the 2d objects. Do I make it clear? Thank you if you could help me with this question.

你好,我在标注自己数据集的时候遇到了同样的问题,请问您的问题解决了吗?