Hello, I see in the paper that the point cloud branch uses an auxiliary detection head, and it significantly improves the results.
Why does it provide such a significant improvement?
Furthermore, why is the auxiliary detection head not used in the fusion model?
Hello, I see in the paper that the point cloud branch uses an auxiliary detection head, and it significantly improves the results. Why does it provide such a significant improvement? Furthermore, why is the auxiliary detection head not used in the fusion model?