search

mileyan / pseudo_lidar

(CVPR 2019) Pseudo-LiDAR from Visual Depth Estimation: Bridging the Gap in 3D Object Detection for Autonomous Driving
https://mileyan.github.io/pseudo_lidar/
MIT License
967 stars 215 forks source link

Pseudo-LiDAR for training AVOD using monodepth2 #40

Open ferdyandannes opened 4 years ago

ferdyandannes commented 4 years ago

Hello First thank you for sharing the amazing works through github, I have several question related with my problem.

  1. I generate the pseudo lidar from your repository by using depth information (NPY files) from monodepth2, but I am not sure whether my pseudo lidar data is correct or not. My pseudo lidar is shown like this: 10_npy_monodepth2 Do you think that my pseudo lidar is correct or not? because when I try to train it using your AVOD repository, it has error in generating the minibatch.

The AVOD error is like this: File "/media/ferdyan/NewDisk/ITRI_3D/avod_pl-master/wavedata/wavedata/tools/core/voxel_grid_2d.py", line 102, in voxelize_2d unique_indices[-1]) IndexError: index -1 is out of bounds for axis 0 with size 0 All Done (Parallel)

Thank you very much. Have a nice day.

mileyan commented 4 years ago

Hi @ferdyandannes , you have to first use the code to generate point cloud. There is a point cloud visualization library pyntcloud. You can use it to visualize Velodyne points and your monocular depth and compare the difference between them.