Hello I am a student studying cloud point based detection and tracking. In your paper, I noticed that a lot of improvements were made with the query voxel. However, there is no mention of how exactly the query voxel is acquired in the paper, and I couldn't find it in the code.
Therefore, I assumed that it is the location of the scores obtained during box prediction. Since the positions of scores above the threshold (0.5) are for low resolution (stride = 8), I multiplied the obtained positions by 8 to visualise them.
This was not the center of the box as mentioned in the paper, but this position is not the position for the input voxel as mentioned in the 3D tracking section.
I'm not going to ask you for the exact code to get the query voxel, but I hope you can at least give me a simple hint as to how to get it.
Hello I am a student studying cloud point based detection and tracking. In your paper, I noticed that a lot of improvements were made with the query voxel. However, there is no mention of how exactly the query voxel is acquired in the paper, and I couldn't find it in the code.
Therefore, I assumed that it is the location of the scores obtained during box prediction. Since the positions of scores above the threshold (0.5) are for low resolution (stride = 8), I multiplied the obtained positions by 8 to visualise them.
This was not the center of the box as mentioned in the paper, but this position is not the position for the input voxel as mentioned in the 3D tracking section.
I'm not going to ask you for the exact code to get the query voxel, but I hope you can at least give me a simple hint as to how to get it.