hailanyi / VirConv

Virtual Sparse Convolution for Multimodal 3D Object Detection
https://arxiv.org/abs/2303.02314
Apache License 2.0
266 stars 39 forks source link

Need some insight into AnchorHead #47

Open FilipaRamos opened 11 months ago

FilipaRamos commented 11 months ago

Hi,

I am trying to understand the box proposals that are generated by the RPN, particularly in the anchor_head_single.py file. I have printed several box predictions and they are quite strange to me. The center coordinates are many times negative but this could make some sense. But the width, height and length also come with negative values!? Could you please help me understand how these proposals could have negative values?

Thanks!