Training for my own custom data about pseudo_lidar HOT 3 OPEN

We are using the PSMNet model here which does disparity prediction on stereo images. If you read the pseudolidar paper linked here: https://arxiv.org/pdf/1812.07179.pdf, you will find that the researchers talk about models that do disparity estimation and pseudo-lidar generation on single images. After getting pseudo-lidar generation from monocular image, you will have to run a 3D object detection model like Frustum PointNet or Sparse ConvNet to get the 3D prediction.

from pseudo_lidar.

@1208overlord,

In our paper, we have used the monocular image to do 3D detection. You can use DORN to generate depth maps first. Then follow step 2.5 to 3 to get 3D predictions. Please let me if you have any questions.

Thank @Hira63S for helping address this issue.

from pseudo_lidar.

Hello, I am going to get 3D bounding box from only one single image. I have been suggested to use this repository for my aim. And then, I have little knowledge about this repository. Could anyone help me with my question? Is this repository suitable for my aim? That is, can I generate 3D bounding box using this repository? If yes, how can I make training dataset from only images dataset and how can I train them? Thank you in advance.

Hello, may I ask how to use single image to generate pseudo-lidar, I found that submission.py need 2 input, imgL and imgR, how to change to single image?
Thank you~

from pseudo_lidar.