Details of the 3D detector

[GehenHe]

Hi Xinshuo,
I enjoyed reading your research work and appreciated for sharing the code. Could you please provide more details of the 3D detector (pointrcnn) that you used? Such as how you train the detector? Which datasets have you used for training?
Much appretiate!

Best wishes.

[WUMINGCHAzero]

I also want to know the details about the detector training. When I use the pre-trained model trained on KITTI Detection Dataset, which is provided by the author in PCDet, I couldn't get the result in your paper. Actually, AP of Cyclist drops a lot. Thanks a lot.

[anti-destiny]

I want more details about the detector too. And my question is about which dataset was used for training your detector. It would be unfair if the whole KITTI dataset including the validation set is used for training the detector, while other researchers do not train their models on the validation set.

[WenyuLWY]

I recommend you follow mmdetection3d. Although the provided models are pre-trained on KITTI 3D Detection benchmark, they perform well on KITTI Tracking Sequences. Second, PointPillars，PointRCNN and Part-A2 are available on KITTI now. Fine-tuning the detection model on KITTI Tracking if needed.

[Tiandooo]

But how can I get the final detection format,whose first column is frame