C++如何切3D矩阵、拼接3D矩阵

最新推荐文章于 2024-07-22 19:51:38 发布

fK0pS

最新推荐文章于 2024-07-22 19:51:38 发布

阅读量430

点赞数 7

文章标签：经验分享

本文链接：https://blog.csdn.net/Hodors/article/details/134660236

版权

C++如何切3D矩阵、拼接3D矩阵

{
    if (mpf_mumap == NULL) {
        std::cout << "Mumap is not loaded!" << std::endl;
        return false;
    }

    //cut mumap into patches
    int i_x = shapes[0];
    int i_y = shapes[1];
    int i_z = shapes[2];
    int i_patch_x = 64;
    int i_patch_y = 64;
    int i_patch_z = 64;
    int i_patch_num_x = i_x / i_patch_x;
    int i_patch_num_y = i_y / i_patch_y;
    int i_patch_num_z = i_z / i_patch_z;
    int i_patch_size = i_patch_x * i_patch_y * i_patch_z;
    float *mpf_patch_mumap = NULL;
    VectorAlloc(mpf_patch_mumap, i_patch_size);
    float *mpf_patch_mask = NULL;
    VectorAlloc(mpf_patch_mask, i_patch_size);
    // infer the mask for each patch
    for (int i = 0; i < i_patch_num_x; i++) {
        for (int j = 0; j < i_patch_num_y; j++) {
            for (int k = 0; k < i_patch_num_z; k++) {
                int i_patch_start_x = i * i_patch_x;
                int i_patch_start_y = j * i_patch_y;
                int i_patch_start_z = k * i_patch_z;
                int i_patch_end_x = i_patch_start_x + i_patch_x;
                int i_patch_end_y = i_patch_start_y + i_patch_y;
                int i_patch_end_z = i_patch_start_z + i_patch_z;
                int i_patch_idx = i * i_patch_num_y * i_patch_num_z + j * i_patch_num_z + k;
                // copy the mumap to mpf_patch_mumap for each pixel
                for (int ii = i_patch_start_x; ii < i_patch_end_x; ii++) {
                    for (int jj = i_patch_start_y; jj < i_patch_end_y; jj++) {
                        for (int kk = i_patch_start_z; kk < i_patch_end_z; kk++) {
//                            int i_mumap_idx = ii * i_y * i_z + jj * i_z + kk; // from copilot
                            int i_mumap_idx = ii * i_x + jj + i_x * i_y * kk; //
                            int i_patch_idx = (ii - i_patch_start_x) * i_patch_x  + (jj - i_patch_start_y)
                                                + i_patch_x * i_patch_y * (kk - i_patch_start_z);
                            mpf_patch_mumap[i_patch_idx] = mpf_mumap[i_mumap_idx];

                        }
                    }
                }
                //infer the mask
                at::Tensor tensor_input = torch::empty({1, 1, i_patch_x, i_patch_y, i_patch_z}, torch::kFloat32);
                memcpy(tensor_input.data_ptr(), mpf_patch_mumap, sizeof(float) * i_patch_size);
                at::Tensor tensor_output = m_module.forward({tensor_input}).toTensor();

                // copy the network output to mpf_mask
                tensor_output = tensor_output.to(torch::kCPU);
                tensor_output = tensor_output.squeeze();
                memcpy(mpf_patch_mask, tensor_output.data_ptr(), sizeof(float) * i_patch_size);

                //update the mpf_mask by mpf_patch_mask
                for (int ii = i_patch_start_x; ii < i_patch_end_x; ii++) {
                    for (int jj = i_patch_start_y; jj < i_patch_end_y; jj++) {
                        for (int kk = i_patch_start_z; kk < i_patch_end_z; kk++) {
                            int i_mask_idx = ii * i_x + jj + i_x * i_y * kk; //
                            int i_patch_idx = (ii - i_patch_start_x) * i_patch_x  + (jj - i_patch_start_y)
                                              + i_patch_x * i_patch_y * (kk - i_patch_start_z);
                            mpf_mask[i_mask_idx] = mpf_patch_mask[i_patch_idx];
                        }
                    }
                }

            }
        }
    }


    return false;

最核心的是这两个代码：

int i_mumap_idx = ii * i_x + jj + i_x * i_y * kk; //
                            int i_patch_idx = (ii - i_patch_start_x) * i_patch_x  + (jj - i_patch_start_y)
                                                + i_patch_x * i_patch_y * (kk - i_patch_start_z);
                            mpf_patch_mumap[i_patch_idx] = mpf_mumap[i_mumap_idx];

对于每个小patch，他的start index就是减去i_patch_start_x。这一点非常的nice