diff --git a/mmcv/ops/csrc/pytorch/npu/ms_deform_attn.cpp b/mmcv/ops/csrc/pytorch/npu/ms_deform_attn.cpp
deleted file mode 100644
index 68ad24d0b41..00000000000
--- a/mmcv/ops/csrc/pytorch/npu/ms_deform_attn.cpp
+++ /dev/null
@@ -1,42 +0,0 @@
-
-#include "pytorch_npu_helper.hpp"
-
-using namespace NPU_NAME_SPACE;
-using namespace std;
-
-Tensor ms_deform_attn_impl_backward_npu(const Tensor &value,
-                                   const Tensor &spatial_shapes,
-                                   const Tensor &level_start_index,
-                                   const Tensor &sampling_loc,
-                                   const Tensor &attn_weight,
-                                   const int im2col_step) {
-    EXEC_NPU_CMD(aclnnMultiScaleDeformableAttnFunction,
-                 value, spatial_shapes, level_start_index, sampling_loc, attn_weight);
-}
-
-void ms_deform_attn_impl_backward_npu(
-    const Tensor &value, const Tensor &spatial_shapes,
-    const Tensor &level_start_index, const Tensor &sampling_loc,
-    const Tensor &attn_weight, const Tensor &grad_output, Tensor &grad_value,
-    Tensor &grad_sampling_loc, Tensor &grad_attn_weight,
-    const int im2col_step) {
-    EXEC_NPU_CMD(aclnnMultiScaleDeformableAttnFunctionGrad,
-                 value, spatial_shapes, level_start_index, sampling_loc, attn_weight,
-                 grad_output, grad_value, grad_sampling_loc, grad_attn_weight);
-}
-
-Tensor ms_deform_attn_impl_backward_npu(const Tensor &value,
-                                        const Tensor &spatial_shapes,
-                                        const Tensor &level_start_index,
-                                        const Tensor &sampling_loc,
-                                        const Tensor &attn_weight,
-                                        const int im2col_step);
-REGISTER_NPU_IMPL(ms_deform_attn_impl_backward, ms_deform_attn_impl_backward_npu);
-
-void ms_deform_attn_impl_backward(const Tensor &value, const Tensor &spatial_shapes,
-                                  const Tensor &level_start_index, const Tensor &sampling_loc,
-                                  const Tensor &attn_weight, const Tensor &grad_output, Tensor &grad_value,
-                                  Tensor &grad_sampling_loc, Tensor &grad_attn_weight,
-                                  const int im2col_step);
-REGISTER_NPU_IMPL(ms_deform_attn_impl_backward,
-                  ms_deform_attn_impl_backward_npu);
diff --git a/mmcv/ops/csrc/pytorch/npu/points_in_box_npu.cpp b/mmcv/ops/csrc/pytorch/npu/points_in_box_npu.cpp
new file mode 100644
index 00000000000..63f0998c69c
--- /dev/null
+++ b/mmcv/ops/csrc/pytorch/npu/points_in_box_npu.cpp
@@ -0,0 +1,18 @@
+#include "pytorch_npu_helper.hpp"
+
+using namespace NPU_NAME_SPACE;
+using namespace std;
+
+void points_in_boxes_part_forward_impl_npu(int batch_size, int boxes_num,
+                                       int pts_num, const Tensor boxes,
+                                       const Tensor pts,
+                                       Tensor box_idx_of_points) {
+    c10::SmallVector<int64_t, 8> output_size = {pts.size(0), pts.size(1)};
+    auto boxes_trans = boxes.transpose(1, 2).contiguous();
+    EXEC_NPU_CMD(aclnnPointsInBox, boxes_trans, pts, box_idx_of_points);
+}
+void points_in_boxes_part_forward_impl(int batch_size, int boxes_num,
+                                       int pts_num, const Tensor boxes,
+                                       const Tensor pts,
+                                       Tensor box_idx_of_points);
+REGISTER_NPU_IMPL(points_in_boxes_part_forward_impl, points_in_boxes_part_forward_impl_npu);