MCG-NJU / SparseOcc

[ECCV 2024] Fully Sparse 3D Occupancy Prediction & RayIoU Evaluation Metric
https://arxiv.org/abs/2312.17118
Apache License 2.0
277 stars 22 forks source link

sparse voxel怎么实现的 #48

Closed tangtaogo closed 3 months ago

tangtaogo commented 3 months ago

你好,感谢你们这么好的开源工作

image

我比较困惑怎么从image feat得到multi-level sparse voxel,哪里的代码对应的是multi-level sparse voxel呢,是用什么数据结构表达的multi-level sparse voxel是octree还是什么形式呢,我看SparseOccHead里面似乎没有找到multi-level sparse voxel在哪里,可以帮忙解答下么?感谢🙏

YANG-CY-163 commented 3 months ago

multi-level sparse voxel部分的代码在models/sparse_voxel_decoder.py里,image feat的交互主要是通过SpaseVoxelDecoder里的SparseBEVSampling和AdaptiveMixing实现,可以结合代码理解一下

tangtaogo commented 3 months ago

感谢回复,所以其实没有显示的生成multi-level sparse voxel,是通过sampling来实现的对嘛

YANG-CY-163 commented 3 months ago

有的,SpaseVoxelDecoder里的query_bbox和query_feat就是和voxel对应的,会经过upsample和sparsify操作(具体可以看models/sparse_voxel_decoder.py),sampling和mixing会更新query_feat, 可以看看SparseBEV了解更多细节

tangtaogo commented 3 months ago

非常感谢