Due to KITTI Semantic datasets' dimension is 4D(xyz+brightness),I tried to imitate a 4d version of chamfer,can you help me to inspect whether my code is correct?
LeopoldACC opened this issue · comments
Hi,@ThibaultGROUEIX Thanks for your work!My 4D version is as below by imitation.
#include <stdio.h>
#include <ATen/ATen.h>
#include <cuda.h>
#include <cuda_runtime.h>
#include <vector>
__global__ void NmDistanceKernel(int b,int n,const float * xyz,int m,const float * xyz2,float * result,int * result_i){
const int batch=2048;
__shared__ float buf[batch*4];
for (int i=blockIdx.x;i<b;i+=gridDim.x){
for (int k2=0;k2<m;k2+=batch){
int end_k=min(m,k2+batch)-k2;
for (int j=threadIdx.x;j<end_k*4;j+=blockDim.x){
buf[j]=xyz2[(i*m+k2)*4+j];
}
__syncthreads();
for (int j=threadIdx.x+blockIdx.y*blockDim.x;j<n;j+=blockDim.x*gridDim.y){
float x1=xyz[(i*n+j)*4+0];
float y1=xyz[(i*n+j)*4+1];
float r1=xyz[(i*n+j)*4+2];
float g1=xyz[(i*n+j)*4+3];
int best_i=0;
float best=0;
int end_ka=end_k-(end_k&4);
if (end_ka==batch){
for (int k=0;k<batch;k+=4){
{
float x2=buf[k*4+0]-x1;
float y2=buf[k*4+1]-y1;
float r2=buf[k*4+2]-r1;
float g2=buf[k*4+3]-g1;
float d=x2*x2+y2*y2+r2*r2+g2*g2;
if (k==0 || d<best){
best=d;
best_i=k+k2;
}
}
{
float x2=buf[k*4+4]-x1;
float y2=buf[k*4+5]-y1;
float r2=buf[k*4+6]-r1;
float g2=buf[k*4+7]-g1;
float d=x2*x2+y2*y2+r2*r2+g2*g2;
if (d<best){
best=d;
best_i=k+k2+1;
}
}
{
float x2=buf[k*4+8]-x1;
float y2=buf[k*4+9]-y1;
float r2=buf[k*4+10]-r1;
float g2=buf[k*4+11]-g1;
float d=x2*x2+y2*y2+r2*r2+g2*g2;
if (d<best){
best=d;
best_i=k+k2+2;
}
}
{
float x2=buf[k*4+12]-x1;
float y2=buf[k*4+13]-y1;
float r2=buf[k*4+14]-r1;
float g2=buf[k*4+15]-g1;
float d=x2*x2+y2*y2+r2*r2+g2*g2;
if (d<best){
best=d;
best_i=k+k2+3;
}
}
}
}else{
for (int k=0;k<end_ka;k+=4){
{
float x2=buf[k*4+0]-x1;
float y2=buf[k*4+1]-y1;
float r2=buf[k*4+2]-r1;
float g2=buf[k*4+3]-g1;
float d=x2*x2+y2*y2+r2*r2+g2*g2;
if (k==0 || d<best){
best=d;
best_i=k+k2;
}
}
{
float x2=buf[k*4+4]-x1;
float y2=buf[k*4+5]-y1;
float r2=buf[k*4+6]-r1;
float g2=buf[k*4+7]-g1;
float d=x2*x2+y2*y2+r2*r2+g2*g2;
if (d<best){
best=d;
best_i=k+k2+1;
}
}
{
float x2=buf[k*4+8]-x1;
float y2=buf[k*4+9]-y1;
float r2=buf[k*4+10]-r1;
float g2=buf[k*4+11]-g1;
float d=x2*x2+y2*y2+r2*r2+g2*g2;
if (d<best){
best=d;
best_i=k+k2+2;
}
}
{
float x2=buf[k*5+12]-x1;
float y2=buf[k*5+13]-y1;
float r2=buf[k*5+14]-r1;
float g2=buf[k*5+15]-g1;
float d=x2*x2+y2*y2+r2*r2+g2*g2;
if (d<best){
best=d;
best_i=k+k2+3;
}
}
}
}
for (int k=end_ka;k<end_k;k++){
float x2=buf[k*5+0]-x1;
float y2=buf[k*5+1]-y1;
float r2=buf[k*5+2]-r1;
float g2=buf[k*5+3]-g1;
float d=x2*x2+y2*y2+r2*r2+g2*g2;
if (k==0 || d<best){
best=d;
best_i=k+k2;
}
}
if (k2==0 || result[(i*n+j)]>best){
result[(i*n+j)]=best;
result_i[(i*n+j)]=best_i;
}
}
__syncthreads();
}
}
}
// int chamfer_cuda_forward(int b,int n,const float * xyz,int m,const float * xyz2,float * result,int * result_i,float * result2,int * result2_i, cudaStream_t stream){
int chamfer_cuda_forward(at::Tensor xyz1, at::Tensor xyz2, at::Tensor dist1, at::Tensor dist2, at::Tensor idx1, at::Tensor idx2){
const auto batch_size = xyz1.size(0);
const auto n = xyz1.size(1); //num_points point cloud A
const auto m = xyz2.size(1); //num_points point cloud B
NmDistanceKernel<<<dim3(32,16,1),512>>>(batch_size, n, xyz1.data<float>(), m, xyz2.data<float>(), dist1.data<float>(), idx1.data<int>());
NmDistanceKernel<<<dim3(32,16,1),512>>>(batch_size, m, xyz2.data<float>(), n, xyz1.data<float>(), dist2.data<float>(), idx2.data<int>());
cudaError_t err = cudaGetLastError();
if (err != cudaSuccess) {
printf("error in nnd updateOutput: %s\n", cudaGetErrorString(err));
//THError("aborting");
return 0;
}
return 1;
}
__global__ void NmDistanceGradKernel(int b,int n,const float * xyz1,int m,const float * xyz2,const float * grad_dist1,const int * idx1,float * grad_xyz1,float * grad_xyz2){
for (int i=blockIdx.x;i<b;i+=gridDim.x){
for (int j=threadIdx.x+blockIdx.y*blockDim.x;j<n;j+=blockDim.x*gridDim.y){
float x1=xyz1[(i*n+j)*5+0];
float y1=xyz1[(i*n+j)*5+1];
float r1=xyz1[(i*n+j)*5+2];
float g1=xyz1[(i*n+j)*5+3];
int j2=idx1[i*n+j];
float x2=xyz2[(i*m+j2)*5+0];
float y2=xyz2[(i*m+j2)*5+1];
float r2=xyz2[(i*m+j2)*5+2];
float g2=xyz2[(i*m+j2)*5+3];
float g=grad_dist1[i*n+j]*2;
atomicAdd(&(grad_xyz1[(i*n+j)*4+0]),g*(x1-x2));
atomicAdd(&(grad_xyz1[(i*n+j)*4+1]),g*(y1-y2));
atomicAdd(&(grad_xyz1[(i*n+j)*4+2]),g*(r1-r2));
atomicAdd(&(grad_xyz1[(i*n+j)*4+3]),g*(g1-g2));
atomicAdd(&(grad_xyz2[(i*m+j2)*4+0]),-(g*(x1-x2)));
atomicAdd(&(grad_xyz2[(i*m+j2)*4+1]),-(g*(y1-y2)));
atomicAdd(&(grad_xyz2[(i*m+j2)*4+2]),-(g*(r1-r2)));
atomicAdd(&(grad_xyz2[(i*m+j2)*4+3]),-(g*(g1-g2)));
}
}
}
// int chamfer_cuda_backward(int b,int n,const float * xyz1,int m,const float * xyz2,const float * grad_dist1,const int * idx1,const float * grad_dist2,const int * idx2,float * grad_xyz1,float * grad_xyz2, cudaStream_t stream){
int chamfer_cuda_backward(at::Tensor xyz1, at::Tensor xyz2, at::Tensor gradxyz1, at::Tensor gradxyz2, at::Tensor graddist1, at::Tensor graddist2, at::Tensor idx1, at::Tensor idx2){
// cudaMemset(grad_xyz1,0,b*n*3*4);
// cudaMemset(grad_xyz2,0,b*m*3*4);
const auto batch_size = xyz1.size(0);
const auto n = xyz1.size(1); //num_points point cloud A
const auto m = xyz2.size(1); //num_points point cloud B
NmDistanceGradKernel<<<dim3(1,16,1),256>>>(batch_size,n,xyz1.data<float>(),m,xyz2.data<float>(),graddist1.data<float>(),idx1.data<int>(),gradxyz1.data<float>(),gradxyz2.data<float>());
NmDistanceGradKernel<<<dim3(1,16,1),256>>>(batch_size,m,xyz2.data<float>(),n,xyz1.data<float>(),graddist2.data<float>(),idx2.data<int>(),gradxyz2.data<float>(),gradxyz1.data<float>());
cudaError_t err = cudaGetLastError();
if (err != cudaSuccess) {
printf("error in nnd get grad: %s\n", cudaGetErrorString(err));
//THError("aborting");
return 0;
}
return 1;
}Another question is there any good tutorial or book to learn how to write cuda op?Thanks a lot!
Looks good ! Sorry I don't know of any good teaching ressource. You could also have used Chamfer5D or 6D and set a few dimensions to 0 in your tensors.
Best regards,