SPengLiang comments

Results 24 comments of


                                            SPengLiang

Some questiones about low_cost mode

你好 1.是的，只有在high模式下，才会使用到人工标签，这个时候可以理解为半监督； 2.low模式没用到3d box标签，因为有标签的话就没必要用low模式了，少量的人工标签就能训练个lidar 检测器来生成较准的pseudo label

depth dense map for test set

Dense depths are only used for training. We do not use depth map for testing, thus we do not generate depth map on test set.

[Errno 2] No such file or directory: './data/kitti/kitti_merge/training/calib_cam2cam/000000.txt'

You can check the dir path (e.g., "./data/kitti/kitti_merge/training/calib_cam2cam", "./data/kitti/kitti_merge" and so on) to locate the path problem.

[Errno 2] No such file or directory: './data/kitti/kitti_merge/training/calib_cam2cam/000000.txt'

Thanks for your feedback! I will check and revise the code.

question about the similar AP in KITTI test set and verification set

It seems that your results are lower than expected, which is weird. We use the default setting in the released code (i.e., 'DLAUp'). Using the released code, some other people...

The difference between depth_dense and the velodyne file in the kitti data set

I am very sorry for the late reply. 1. "depth_dense" provides dense depth information by employing depth completion on sparse depth maps (namely, from kitti.velodyne) while "kitti.velodyne" contains only sparse...

The Ap of batchsize 16 in one gpu

Sorry for the late reply. You can increase the learning rate and the training epochs, and the learning rate decay strategy should be adjusted correspondingly.

hello! i want know what is the depth_dense?how can i get it?

You can use a pre-trained depth completion model (https://github.com/JUGGHM/PENet_ICRA2021) to generate dense depth maps. To ease the usage, we provided the pre-generated depth maps (https://drive.google.com/file/d/1mlHtG8ZXLfjm0lSpUOXHulGF9fsthRtM/view?usp=sharing) in Readme. You can also...

d = cv.imread('{}/{:0>6}.png'.format(self.dense_depth_dir, index), -1) / 256.

深度图保存时是灰度图，为了精度是先乘256再存成uint16。code里面opencv读取灰度图的那种方式是需要除以256.来获取真实的深度的

Support for Waymo

Sorry for the late reply! Thank you very much for your interest in this work. Similar to CaDDN, we convert the Waymo data format to the same format as KITTI....