Questions for the class prediction and label annotation about coperception

[flow233-ai]

I got it. Thanks again for your meticulous answer！

[dekunma]

For your first question, I think @RoboticsYimingLi could probably give a more precise answer.

For your second question, as the variable name suggests, the labels are one-hot encoded, so the last two dimensions are one-hot labels for each voxel for each agent.
For example, label_one_hot[0][0][0] could be

[[1, 0],
[1, 0],
[1, 0],
[1, 0],
[1, 0],
[1, 0]]

The first dimension 6 means there are 6 agents in the dataset, and the second dimension 2 means there are 2 categories, as specified in

coperception/coperception/configs/Config.py

Lines 172 to 173 in 7e99052

	if self.binary:
	self.category_num = 2

So, it means that the label for agent 0 at its voxelized birds-eye-view point cloud position [0, 0] is 0.

[flow233-ai]

Thanks for your answer.
As for the first question, I will contact https://github.com/RoboticsYimingLi for more precise answer.
But for the second question, I am still confused. As I mentioned above, the shape of variable label_one_hot is [6,256,256,6,2]. If the foruth dimenstion means there are 6 agents in the dataset, what does the first dimenstion mean? You mean they are both standing for 6 agents in the dataset?
I found that for the fourth dimension of label_one_hot[0][0][0], the six elements are always the same.

[dekunma]

The first dimension actually means the number of agents involved in the current training process. If you set rsu to 0 to exclude RSU, then it will be 5.

---

the six elements are always the same.

Yes, they should always be the same because the ground truth for a voxel should be the same for all agents.

[simbaforrest]

@flow233-ai Is your issue addressed? If so, please close it. Thank you.

[dekunma]

@flow233-ai Since this issue seems to be resolved, I'm closing it now. Feel free to reopen it if you have any follow-up questions.