Closed Shahmad99 closed 1 year ago
Hello, I appreciate your interest in our project.
The way I understood your question is that you want to use the projection module so that you can transform point cloud into an image-like feature map.
This can be performed using the projection module in the projection branch, which is implemented as part of the projection-based convolutional module in the following code. https://github.com/pahn04/PPConv/blob/395957b919786bb5b603f37a94ccf9173afce085/pvcnn_code/modules/conv_pillar_caf.py#L89 The output variable proj_x is in the form of B x C x H x W, where B is the batch size and the other dimensions denote the size of pseudo-image feature map. The projection module (which is called as self.projection) is implemented in the following code. https://github.com/pahn04/PPConv/blob/395957b919786bb5b603f37a94ccf9173afce085/pvcnn_code/modules/projection.py
For further explanation, you can refer to the section IV. B. 1) in our paper.
Hi, can we use the just "projection network" only to generate pseudo image from point cloud? i want to perform some other experiment on pseudo image. But I am confused how to use only the projection network from this code. Thanks