FPS(Farthest Point Sampling) 编译笔记

目录

作者自己实现了一版：

c代码：

FPS(Farthest Point Sampling)--最远点采样算法

报错代码：

报错解决方法：

---

作者自己实现了一版：

extension-cpp/cuda/setup.py at dbf58c505fe5035eccdb71195b6e49c6694f7bbc · hetong007/extension-cpp · GitHub

c代码：

https://github.com/Barbany/fps_cuda/tree/main

FPS(Farthest Point Sampling)--最远点采样算法

新版代码：

pointnet2_ops_lib/pointnet2_ops_lib/pointnet2_ops/pointnet2_utils.py at main · ybhhdt/pointnet2_ops_lib · GitHub

class BallQuery(Function):
    @staticmethod
    def forward(ctx, radius, nsample, xyz, new_xyz):
        # type: (Any, float, int, torch.Tensor, torch.Tensor) -> torch.Tensor
        r"""

        Parameters
        ----------
        radius : float
            radius of the balls
        nsample : int
            maximum number of features in the balls
        xyz : torch.Tensor
            (B, N, 3) xyz coordinates of the features
        new_xyz : torch.Tensor
            (B, npoint, 3) centers of the ball query

        Returns
        -------
        torch.Tensor
            (B, npoint, nsample) tensor with the indicies of the features that form the query balls
        """
        idx, pts_cnt= _ext.ball_query(new_xyz, xyz, radius, nsample)

        ctx.mark_non_differentiable(idx)
        ctx.mark_non_differentiable(pts_cnt)

        return idx, pts_cnt

报错代码：

 
 
import torch
from torch.autograd import Function
 
try:
    import builtins
except:
    import __builtin__ as builtins
 
try:
    import fps_cuda._ext as _ext
except ImportError:
    if not getattr(builtins, "__POINTNET2_SETUP__", False):
        raise ImportError(
            "Could not import _ext module.\n"
            "Please see the setup instructions in the README: "
            "https://github.com/erikwijmans/Pointnet2_PyTorch/blob/master/README.rst"
        )
 
class FurthestPointSampling(Function):
    @staticmethod
    def forward(ctx, xyz, npoint):
        # type: (Any, torch.Tensor, int) -> torch.Tensor
        r"""
        Uses iterative furthest point sampling to select a set of npoint features that have the largest
        minimum distance
        Parameters
        ----------
        xyz : torch.Tensor
            (B, N, 3) tensor where N > npoint
        npoint : int32
            number of features in the sampled set
        Returns
        -------
        torch.Tensor
            (B, npoint) tensor containing the set
        """
        fps_inds = _ext.furthest_point_sampling(xyz, npoint)
        ctx.mark_non_differentiable(fps_inds)
        return fps_inds
 
    @staticmethod
    def backward(xyz, a=None):
        return None, None
 
 
furthest_point_sample = FurthestPointSampling.apply    
 
 
xyz = torch.randn(5,1024,3)
npoint=512
fps_inds = furthest_point_sample(xyz, npoint)
# ctx.mark_non_differentiable(fps_inds)
print(fps_inds)

RuntimeError: Unknown layout in MultiScaleDeformableAttnFunction · Issue #284 · IDEA-Research/GroundingDINO · GitHub

报错解决方法：

pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu118