diff --git a/test/tracing/frcnn/test_frcnn_tracing.cpp b/test/tracing/frcnn/test_frcnn_tracing.cpp
index 90476b24f4b..7b10aee3c89 100644
--- a/test/tracing/frcnn/test_frcnn_tracing.cpp
+++ b/test/tracing/frcnn/test_frcnn_tracing.cpp
@@ -7,7 +7,7 @@
 #ifdef _WIN32
 // Windows only
 // This is necessary until operators are automatically registered on include
-static auto _nms = &nms_cpu;
+static auto _nms = &vision::ops::nms_cpu;
 #endif
 
 int main() {
diff --git a/torchvision/csrc/cpu/deform_conv2d_kernel.cpp b/torchvision/csrc/cpu/deform_conv2d_kernel.cpp
index 5cac99db04a..4ae2d0a02db 100644
--- a/torchvision/csrc/cpu/deform_conv2d_kernel.cpp
+++ b/torchvision/csrc/cpu/deform_conv2d_kernel.cpp
@@ -68,6 +68,9 @@
 
 #include "deform_conv2d_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 const int kMaxParallelImgs = 32;
@@ -1137,3 +1140,6 @@ deform_conv2d_backward_cpu(
   return std::make_tuple(
       grad_input, grad_weight, grad_offset, grad_mask, grad_bias);
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cpu/deform_conv2d_kernel.h b/torchvision/csrc/cpu/deform_conv2d_kernel.h
index 2eb5ab37c6e..2a49bad8304 100644
--- a/torchvision/csrc/cpu/deform_conv2d_kernel.h
+++ b/torchvision/csrc/cpu/deform_conv2d_kernel.h
@@ -3,6 +3,9 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API at::Tensor deform_conv2d_forward_cpu(
     const at::Tensor& input,
     const at::Tensor& weight,
@@ -37,3 +40,6 @@ VISION_API std::
         int64_t n_weight_grps,
         int64_t n_offset_grps,
         bool use_mask);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cpu/nms_kernel.cpp b/torchvision/csrc/cpu/nms_kernel.cpp
index 52953a9e822..a77a6906870 100644
--- a/torchvision/csrc/cpu/nms_kernel.cpp
+++ b/torchvision/csrc/cpu/nms_kernel.cpp
@@ -1,5 +1,8 @@
 #include "nms_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 template <typename scalar_t>
@@ -103,3 +106,6 @@ at::Tensor nms_cpu(
   });
   return result;
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cpu/nms_kernel.h b/torchvision/csrc/cpu/nms_kernel.h
index 7b6ef442626..1fdcaf3d3f9 100644
--- a/torchvision/csrc/cpu/nms_kernel.h
+++ b/torchvision/csrc/cpu/nms_kernel.h
@@ -3,7 +3,13 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API at::Tensor nms_cpu(
     const at::Tensor& dets,
     const at::Tensor& scores,
     double iou_threshold);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cpu/ps_roi_align_kernel.cpp b/torchvision/csrc/cpu/ps_roi_align_kernel.cpp
index 3d6c95f02ea..5abe4a41477 100644
--- a/torchvision/csrc/cpu/ps_roi_align_kernel.cpp
+++ b/torchvision/csrc/cpu/ps_roi_align_kernel.cpp
@@ -1,5 +1,8 @@
 #include "ps_roi_align_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 template <typename T>
@@ -416,3 +419,6 @@ at::Tensor ps_roi_align_backward_cpu(
       });
   return grad_input;
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cpu/ps_roi_align_kernel.h b/torchvision/csrc/cpu/ps_roi_align_kernel.h
index 86a3f9a8876..a4bea77853b 100644
--- a/torchvision/csrc/cpu/ps_roi_align_kernel.h
+++ b/torchvision/csrc/cpu/ps_roi_align_kernel.h
@@ -3,6 +3,9 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API std::tuple<at::Tensor, at::Tensor> ps_roi_align_forward_cpu(
     const at::Tensor& input,
     const at::Tensor& rois,
@@ -23,3 +26,6 @@ VISION_API at::Tensor ps_roi_align_backward_cpu(
     int64_t channels,
     int64_t height,
     int64_t width);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cpu/ps_roi_pool_kernel.cpp b/torchvision/csrc/cpu/ps_roi_pool_kernel.cpp
index cdee9b9f55c..425b4c68f1a 100644
--- a/torchvision/csrc/cpu/ps_roi_pool_kernel.cpp
+++ b/torchvision/csrc/cpu/ps_roi_pool_kernel.cpp
@@ -1,5 +1,8 @@
 #include "ps_roi_pool_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 template <class T>
@@ -255,3 +258,6 @@ at::Tensor ps_roi_pool_backward_cpu(
       });
   return grad_input;
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cpu/ps_roi_pool_kernel.h b/torchvision/csrc/cpu/ps_roi_pool_kernel.h
index 14a4e22681a..2cefe39e11e 100644
--- a/torchvision/csrc/cpu/ps_roi_pool_kernel.h
+++ b/torchvision/csrc/cpu/ps_roi_pool_kernel.h
@@ -3,6 +3,9 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API std::tuple<at::Tensor, at::Tensor> ps_roi_pool_forward_cpu(
     const at::Tensor& input,
     const at::Tensor& rois,
@@ -21,3 +24,6 @@ VISION_API at::Tensor ps_roi_pool_backward_cpu(
     int64_t channels,
     int64_t height,
     int64_t width);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cpu/roi_align_kernel.cpp b/torchvision/csrc/cpu/roi_align_kernel.cpp
index 133722fdc5e..cbb75f2c474 100644
--- a/torchvision/csrc/cpu/roi_align_kernel.cpp
+++ b/torchvision/csrc/cpu/roi_align_kernel.cpp
@@ -1,5 +1,8 @@
 #include "roi_align_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 // implementation taken from Caffe2
@@ -494,3 +497,6 @@ at::Tensor roi_align_backward_cpu(
       });
   return grad_input;
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cpu/roi_align_kernel.h b/torchvision/csrc/cpu/roi_align_kernel.h
index 79fd46bd44e..2e7813c261c 100644
--- a/torchvision/csrc/cpu/roi_align_kernel.h
+++ b/torchvision/csrc/cpu/roi_align_kernel.h
@@ -3,6 +3,9 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API at::Tensor roi_align_forward_cpu(
     const at::Tensor& input,
     const at::Tensor& rois,
@@ -24,3 +27,6 @@ VISION_API at::Tensor roi_align_backward_cpu(
     int64_t width,
     int64_t sampling_ratio,
     bool aligned);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cpu/roi_pool_kernel.cpp b/torchvision/csrc/cpu/roi_pool_kernel.cpp
index d622f2b430b..375b722684e 100644
--- a/torchvision/csrc/cpu/roi_pool_kernel.cpp
+++ b/torchvision/csrc/cpu/roi_pool_kernel.cpp
@@ -2,6 +2,9 @@
 
 #include "roi_pool_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 template <class T>
@@ -231,3 +234,6 @@ at::Tensor roi_pool_backward_cpu(
       });
   return grad_input;
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cpu/roi_pool_kernel.h b/torchvision/csrc/cpu/roi_pool_kernel.h
index 66fd993d5b4..33d029cf31a 100644
--- a/torchvision/csrc/cpu/roi_pool_kernel.h
+++ b/torchvision/csrc/cpu/roi_pool_kernel.h
@@ -3,6 +3,9 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API std::tuple<at::Tensor, at::Tensor> roi_pool_forward_cpu(
     const at::Tensor& input,
     const at::Tensor& rois,
@@ -21,3 +24,6 @@ VISION_API at::Tensor roi_pool_backward_cpu(
     int64_t channels,
     int64_t height,
     int64_t width);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/cuda_helpers.h b/torchvision/csrc/cuda/cuda_helpers.h
index a82e1c86f87..cec4a183899 100644
--- a/torchvision/csrc/cuda/cuda_helpers.h
+++ b/torchvision/csrc/cuda/cuda_helpers.h
@@ -1,5 +1,8 @@
 #pragma once
 
+namespace vision {
+namespace ops {
+
 #define CUDA_1D_KERNEL_LOOP(i, n)                                \
   for (int i = (blockIdx.x * blockDim.x) + threadIdx.x; i < (n); \
        i += (blockDim.x * gridDim.x))
@@ -8,3 +11,6 @@ template <typename integer>
 constexpr __host__ __device__ inline integer ceil_div(integer n, integer m) {
   return (n + m - 1) / m;
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/deform_conv2d_kernel.cu b/torchvision/csrc/cuda/deform_conv2d_kernel.cu
index cef8124caf3..e530710863c 100644
--- a/torchvision/csrc/cuda/deform_conv2d_kernel.cu
+++ b/torchvision/csrc/cuda/deform_conv2d_kernel.cu
@@ -73,6 +73,9 @@
 #include "cuda_helpers.h"
 #include "deform_conv2d_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 const int kMaxParallelImgs = 32;
@@ -1183,3 +1186,6 @@ deform_conv2d_backward_cuda(
   return std::make_tuple(
       grad_input, grad_weight, grad_offset, grad_mask, grad_bias);
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/deform_conv2d_kernel.h b/torchvision/csrc/cuda/deform_conv2d_kernel.h
index 00f3f3dc15d..b2e3dc3f17f 100644
--- a/torchvision/csrc/cuda/deform_conv2d_kernel.h
+++ b/torchvision/csrc/cuda/deform_conv2d_kernel.h
@@ -3,6 +3,9 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API at::Tensor deform_conv2d_forward_cuda(
     const at::Tensor& input,
     const at::Tensor& weight,
@@ -37,3 +40,6 @@ VISION_API std::
         int64_t n_weight_grps,
         int64_t n_offset_grps,
         bool use_mask);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/nms_kernel.cu b/torchvision/csrc/cuda/nms_kernel.cu
index ae244efebe7..b8d4b3ce0ec 100644
--- a/torchvision/csrc/cuda/nms_kernel.cu
+++ b/torchvision/csrc/cuda/nms_kernel.cu
@@ -4,6 +4,9 @@
 #include "cuda_helpers.h"
 #include "nms_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 int const threadsPerBlock = sizeof(unsigned long long) * 8;
@@ -162,3 +165,6 @@ at::Tensor nms_cuda(
       {keep.narrow(/*dim=*/0, /*start=*/0, /*length=*/num_to_keep)
            .to(order_t.device(), keep.scalar_type())});
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/nms_kernel.h b/torchvision/csrc/cuda/nms_kernel.h
index 1eceddaccf3..0d2c0838437 100644
--- a/torchvision/csrc/cuda/nms_kernel.h
+++ b/torchvision/csrc/cuda/nms_kernel.h
@@ -3,7 +3,13 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API at::Tensor nms_cuda(
     const at::Tensor& dets,
     const at::Tensor& scores,
     double iou_threshold);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/ps_roi_align_kernel.cu b/torchvision/csrc/cuda/ps_roi_align_kernel.cu
index 7c808580258..6b1e729b12d 100644
--- a/torchvision/csrc/cuda/ps_roi_align_kernel.cu
+++ b/torchvision/csrc/cuda/ps_roi_align_kernel.cu
@@ -5,6 +5,9 @@
 #include "cuda_helpers.h"
 #include "ps_roi_align_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 template <typename T>
@@ -434,3 +437,6 @@ at::Tensor ps_roi_align_backward_cuda(
   AT_CUDA_CHECK(cudaGetLastError());
   return grad_input;
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/ps_roi_align_kernel.h b/torchvision/csrc/cuda/ps_roi_align_kernel.h
index 45a300d6711..c40e6fa55b1 100644
--- a/torchvision/csrc/cuda/ps_roi_align_kernel.h
+++ b/torchvision/csrc/cuda/ps_roi_align_kernel.h
@@ -3,6 +3,9 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API std::tuple<at::Tensor, at::Tensor> ps_roi_align_forward_cuda(
     const at::Tensor& input,
     const at::Tensor& rois,
@@ -23,3 +26,6 @@ VISION_API at::Tensor ps_roi_align_backward_cuda(
     int64_t channels,
     int64_t height,
     int64_t width);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/ps_roi_pool_kernel.cu b/torchvision/csrc/cuda/ps_roi_pool_kernel.cu
index ed0ed26484d..91fd25b4bb5 100644
--- a/torchvision/csrc/cuda/ps_roi_pool_kernel.cu
+++ b/torchvision/csrc/cuda/ps_roi_pool_kernel.cu
@@ -5,6 +5,9 @@
 #include "cuda_helpers.h"
 #include "ps_roi_pool_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 template <typename T>
@@ -270,3 +273,6 @@ at::Tensor ps_roi_pool_backward_cuda(
   AT_CUDA_CHECK(cudaGetLastError());
   return grad_input;
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/ps_roi_pool_kernel.h b/torchvision/csrc/cuda/ps_roi_pool_kernel.h
index e97f0ee7065..21015d4693b 100644
--- a/torchvision/csrc/cuda/ps_roi_pool_kernel.h
+++ b/torchvision/csrc/cuda/ps_roi_pool_kernel.h
@@ -3,6 +3,9 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API std::tuple<at::Tensor, at::Tensor> ps_roi_pool_forward_cuda(
     const at::Tensor& input,
     const at::Tensor& rois,
@@ -21,3 +24,6 @@ VISION_API at::Tensor ps_roi_pool_backward_cuda(
     int64_t channels,
     int64_t height,
     int64_t width);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/roi_align_kernel.cu b/torchvision/csrc/cuda/roi_align_kernel.cu
index 195d8b067f4..59388faa6ad 100644
--- a/torchvision/csrc/cuda/roi_align_kernel.cu
+++ b/torchvision/csrc/cuda/roi_align_kernel.cu
@@ -5,6 +5,9 @@
 #include "cuda_helpers.h"
 #include "roi_align_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 template <typename T>
@@ -443,3 +446,6 @@ at::Tensor roi_align_backward_cuda(
   AT_CUDA_CHECK(cudaGetLastError());
   return grad_input;
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/roi_align_kernel.h b/torchvision/csrc/cuda/roi_align_kernel.h
index 46054f04f38..71096201627 100644
--- a/torchvision/csrc/cuda/roi_align_kernel.h
+++ b/torchvision/csrc/cuda/roi_align_kernel.h
@@ -3,6 +3,9 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API at::Tensor roi_align_forward_cuda(
     const at::Tensor& input,
     const at::Tensor& rois,
@@ -24,3 +27,6 @@ VISION_API at::Tensor roi_align_backward_cuda(
     int64_t width,
     int64_t sampling_ratio,
     bool aligned);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/roi_pool_kernel.cu b/torchvision/csrc/cuda/roi_pool_kernel.cu
index 782ecaf9eb3..a96e79c87a9 100644
--- a/torchvision/csrc/cuda/roi_pool_kernel.cu
+++ b/torchvision/csrc/cuda/roi_pool_kernel.cu
@@ -6,6 +6,9 @@
 #include "cuda_helpers.h"
 #include "roi_pool_kernel.h"
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 template <typename T>
@@ -254,3 +257,6 @@ at::Tensor roi_pool_backward_cuda(
   AT_CUDA_CHECK(cudaGetLastError());
   return grad_input;
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/cuda/roi_pool_kernel.h b/torchvision/csrc/cuda/roi_pool_kernel.h
index 3a99f7521bd..71a649968db 100644
--- a/torchvision/csrc/cuda/roi_pool_kernel.h
+++ b/torchvision/csrc/cuda/roi_pool_kernel.h
@@ -3,6 +3,9 @@
 #include <ATen/ATen.h>
 #include "../macros.h"
 
+namespace vision {
+namespace ops {
+
 VISION_API std::tuple<at::Tensor, at::Tensor> roi_pool_forward_cuda(
     const at::Tensor& input,
     const at::Tensor& rois,
@@ -21,3 +24,6 @@ VISION_API at::Tensor roi_pool_backward_cuda(
     int64_t channels,
     int64_t height,
     int64_t width);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/deform_conv2d.cpp b/torchvision/csrc/deform_conv2d.cpp
index 74ba630537a..e8a416683f2 100644
--- a/torchvision/csrc/deform_conv2d.cpp
+++ b/torchvision/csrc/deform_conv2d.cpp
@@ -5,6 +5,9 @@
 #include <ATen/autocast_mode.h>
 #endif
 
+namespace vision {
+namespace ops {
+
 at::Tensor deform_conv2d(
     const at::Tensor& input,
     const at::Tensor& weight,
@@ -361,3 +364,6 @@ deform_conv2d_backward_autograd(
 
   return std::make_tuple(result[0], result[1], result[2], result[3], result[4]);
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/deform_conv2d.h b/torchvision/csrc/deform_conv2d.h
index 6adc77fb888..85675ee6bf2 100644
--- a/torchvision/csrc/deform_conv2d.h
+++ b/torchvision/csrc/deform_conv2d.h
@@ -9,6 +9,9 @@
 #include "hip/deform_conv2d_kernel.h"
 #endif
 
+namespace vision {
+namespace ops {
+
 // C++ Forward
 at::Tensor deform_conv2d(
     const at::Tensor& input,
@@ -98,3 +101,6 @@ deform_conv2d_backward_autograd(
     int64_t groups,
     int64_t offset_groups,
     bool use_mask);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/models/alexnet.h b/torchvision/csrc/models/alexnet.h
index 33ffe379a97..d2529c88882 100644
--- a/torchvision/csrc/models/alexnet.h
+++ b/torchvision/csrc/models/alexnet.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/torch.h>
-#include "general.h"
+#include "../macros.h"
 
 namespace vision {
 namespace models {
diff --git a/torchvision/csrc/models/densenet.h b/torchvision/csrc/models/densenet.h
index 22db45b719d..9358631e4ba 100644
--- a/torchvision/csrc/models/densenet.h
+++ b/torchvision/csrc/models/densenet.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/torch.h>
-#include "general.h"
+#include "../macros.h"
 
 namespace vision {
 namespace models {
diff --git a/torchvision/csrc/models/general.h b/torchvision/csrc/models/general.h
deleted file mode 100644
index 4463786b4bb..00000000000
--- a/torchvision/csrc/models/general.h
+++ /dev/null
@@ -1,11 +0,0 @@
-#pragma once
-
-#ifdef _WIN32
-#if defined(torchvision_EXPORTS)
-#define VISION_API __declspec(dllexport)
-#else
-#define VISION_API __declspec(dllimport)
-#endif
-#else
-#define VISION_API
-#endif
diff --git a/torchvision/csrc/models/googlenet.h b/torchvision/csrc/models/googlenet.h
index d5192c7623f..3d4b3faf7e4 100644
--- a/torchvision/csrc/models/googlenet.h
+++ b/torchvision/csrc/models/googlenet.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/torch.h>
-#include "general.h"
+#include "../macros.h"
 
 namespace vision {
 namespace models {
diff --git a/torchvision/csrc/models/inception.h b/torchvision/csrc/models/inception.h
index 3f964e3103c..08f329fbc1b 100644
--- a/torchvision/csrc/models/inception.h
+++ b/torchvision/csrc/models/inception.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/torch.h>
-#include "general.h"
+#include "../macros.h"
 
 namespace vision {
 namespace models {
diff --git a/torchvision/csrc/models/mnasnet.h b/torchvision/csrc/models/mnasnet.h
index ae136cd5b30..e69559b2dd7 100644
--- a/torchvision/csrc/models/mnasnet.h
+++ b/torchvision/csrc/models/mnasnet.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/torch.h>
-#include "general.h"
+#include "../macros.h"
 
 namespace vision {
 namespace models {
diff --git a/torchvision/csrc/models/mobilenet.h b/torchvision/csrc/models/mobilenet.h
index 7e3f8596692..63d26474b23 100644
--- a/torchvision/csrc/models/mobilenet.h
+++ b/torchvision/csrc/models/mobilenet.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/torch.h>
-#include "general.h"
+#include "../macros.h"
 
 namespace vision {
 namespace models {
diff --git a/torchvision/csrc/models/resnet.h b/torchvision/csrc/models/resnet.h
index ee4a8cd5284..b8caf4332c6 100644
--- a/torchvision/csrc/models/resnet.h
+++ b/torchvision/csrc/models/resnet.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/torch.h>
-#include "general.h"
+#include "../macros.h"
 
 namespace vision {
 namespace models {
diff --git a/torchvision/csrc/models/shufflenetv2.h b/torchvision/csrc/models/shufflenetv2.h
index 8aa0ea69db4..ea8524c72fa 100644
--- a/torchvision/csrc/models/shufflenetv2.h
+++ b/torchvision/csrc/models/shufflenetv2.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/torch.h>
-#include "general.h"
+#include "../macros.h"
 
 namespace vision {
 namespace models {
diff --git a/torchvision/csrc/models/squeezenet.h b/torchvision/csrc/models/squeezenet.h
index 5cd3cb482e7..c43d2b07f13 100644
--- a/torchvision/csrc/models/squeezenet.h
+++ b/torchvision/csrc/models/squeezenet.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/torch.h>
-#include "general.h"
+#include "../macros.h"
 
 namespace vision {
 namespace models {
diff --git a/torchvision/csrc/models/vgg.h b/torchvision/csrc/models/vgg.h
index ea64dbfa151..4a540bea822 100644
--- a/torchvision/csrc/models/vgg.h
+++ b/torchvision/csrc/models/vgg.h
@@ -1,7 +1,7 @@
 #pragma once
 
 #include <torch/torch.h>
-#include "general.h"
+#include "../macros.h"
 
 namespace vision {
 namespace models {
diff --git a/torchvision/csrc/new_empty_tensor_op.cpp b/torchvision/csrc/new_empty_tensor_op.cpp
index e4f31600c54..768b5599bf9 100644
--- a/torchvision/csrc/new_empty_tensor_op.cpp
+++ b/torchvision/csrc/new_empty_tensor_op.cpp
@@ -3,6 +3,9 @@
 #include "new_empty_tensor_op.h"
 #include <torch/extension.h>
 
+namespace vision {
+namespace ops {
+
 namespace {
 
 class NewEmptyTensorOp : public torch::autograd::Function<NewEmptyTensorOp> {
@@ -33,3 +36,6 @@ at::Tensor new_empty_tensor(
     const c10::List<int64_t>& shape) {
   return NewEmptyTensorOp::apply(input, shape)[0];
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/new_empty_tensor_op.h b/torchvision/csrc/new_empty_tensor_op.h
index 75f4cd5a7fe..f00cb67b779 100644
--- a/torchvision/csrc/new_empty_tensor_op.h
+++ b/torchvision/csrc/new_empty_tensor_op.h
@@ -2,6 +2,12 @@
 
 #include <ATen/ATen.h>
 
+namespace vision {
+namespace ops {
+
 at::Tensor new_empty_tensor(
     const at::Tensor& input,
     const c10::List<int64_t>& shape);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/nms.cpp b/torchvision/csrc/nms.cpp
index 075f3101937..2f9dbee9a32 100644
--- a/torchvision/csrc/nms.cpp
+++ b/torchvision/csrc/nms.cpp
@@ -5,6 +5,9 @@
 #include <ATen/autocast_mode.h>
 #endif
 
+namespace vision {
+namespace ops {
+
 at::Tensor nms(
     const at::Tensor& dets,
     const at::Tensor& scores,
@@ -27,3 +30,6 @@ at::Tensor nms_autocast(
       iou_threshold);
 }
 #endif
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/nms.h b/torchvision/csrc/nms.h
index 87b07548454..ac7cbc53caf 100644
--- a/torchvision/csrc/nms.h
+++ b/torchvision/csrc/nms.h
@@ -9,6 +9,9 @@
 #include "hip/nms_kernel.h"
 #endif
 
+namespace vision {
+namespace ops {
+
 // C++ Forward
 at::Tensor nms(
     const at::Tensor& dets,
@@ -22,3 +25,6 @@ at::Tensor nms_autocast(
     const at::Tensor& scores,
     double iou_threshold);
 #endif
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/ps_roi_align.cpp b/torchvision/csrc/ps_roi_align.cpp
index 0e1a30d6e63..5add21aaeec 100644
--- a/torchvision/csrc/ps_roi_align.cpp
+++ b/torchvision/csrc/ps_roi_align.cpp
@@ -5,6 +5,9 @@
 #include <ATen/autocast_mode.h>
 #endif
 
+namespace vision {
+namespace ops {
+
 std::tuple<at::Tensor, at::Tensor> ps_roi_align(
     const at::Tensor& input,
     const at::Tensor& rois,
@@ -218,3 +221,6 @@ at::Tensor ps_roi_align_backward_autograd(
       height,
       width)[0];
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/ps_roi_align.h b/torchvision/csrc/ps_roi_align.h
index 0f7ecea2f12..c21107df4f4 100644
--- a/torchvision/csrc/ps_roi_align.h
+++ b/torchvision/csrc/ps_roi_align.h
@@ -9,6 +9,9 @@
 #include "hip/ps_roi_align_kernel.h"
 #endif
 
+namespace vision {
+namespace ops {
+
 // C++ Forward
 std::tuple<at::Tensor, at::Tensor> ps_roi_align(
     const at::Tensor& input,
@@ -64,3 +67,6 @@ at::Tensor ps_roi_align_backward_autograd(
     int64_t channels,
     int64_t height,
     int64_t width);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/ps_roi_pool.cpp b/torchvision/csrc/ps_roi_pool.cpp
index 76fb2d04be7..88a733a6369 100644
--- a/torchvision/csrc/ps_roi_pool.cpp
+++ b/torchvision/csrc/ps_roi_pool.cpp
@@ -5,6 +5,9 @@
 #include <ATen/autocast_mode.h>
 #endif
 
+namespace vision {
+namespace ops {
+
 std::tuple<at::Tensor, at::Tensor> ps_roi_pool(
     const at::Tensor& input,
     const at::Tensor& rois,
@@ -197,3 +200,6 @@ at::Tensor ps_roi_pool_backward_autograd(
       height,
       width)[0];
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/ps_roi_pool.h b/torchvision/csrc/ps_roi_pool.h
index 0c8baef4a9a..20ae17d3ad1 100644
--- a/torchvision/csrc/ps_roi_pool.h
+++ b/torchvision/csrc/ps_roi_pool.h
@@ -9,6 +9,9 @@
 #include "hip/ps_roi_pool_kernel.h"
 #endif
 
+namespace vision {
+namespace ops {
+
 // C++ Forward
 std::tuple<at::Tensor, at::Tensor> ps_roi_pool(
     const at::Tensor& input,
@@ -59,3 +62,6 @@ at::Tensor ps_roi_pool_backward_autograd(
     int64_t channels,
     int64_t height,
     int64_t width);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/roi_align.cpp b/torchvision/csrc/roi_align.cpp
index 30eda8612d2..63643a6cb46 100644
--- a/torchvision/csrc/roi_align.cpp
+++ b/torchvision/csrc/roi_align.cpp
@@ -5,6 +5,9 @@
 #include <ATen/autocast_mode.h>
 #endif
 
+namespace vision {
+namespace ops {
+
 at::Tensor roi_align(
     const at::Tensor& input, // Input feature map.
     const at::Tensor& rois, // List of ROIs to pool over.
@@ -227,3 +230,6 @@ at::Tensor roi_align_backward_autograd(
       sampling_ratio,
       aligned)[0];
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/roi_align.h b/torchvision/csrc/roi_align.h
index d9bae4ba2a1..1e92c8d2134 100644
--- a/torchvision/csrc/roi_align.h
+++ b/torchvision/csrc/roi_align.h
@@ -9,6 +9,9 @@
 #include "hip/roi_align_kernel.h"
 #endif
 
+namespace vision {
+namespace ops {
+
 // C++ Forward
 at::Tensor roi_align(
     const at::Tensor& input,
@@ -67,3 +70,6 @@ at::Tensor roi_align_backward_autograd(
     int64_t width,
     int64_t sampling_ratio,
     bool aligned);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/roi_pool.cpp b/torchvision/csrc/roi_pool.cpp
index c8d70bd8940..b2948e6dd23 100644
--- a/torchvision/csrc/roi_pool.cpp
+++ b/torchvision/csrc/roi_pool.cpp
@@ -5,6 +5,9 @@
 #include <ATen/autocast_mode.h>
 #endif
 
+namespace vision {
+namespace ops {
+
 std::tuple<at::Tensor, at::Tensor> roi_pool(
     const at::Tensor& input,
     const at::Tensor& rois,
@@ -196,3 +199,6 @@ at::Tensor roi_pool_backward_autograd(
       height,
       width)[0];
 }
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/roi_pool.h b/torchvision/csrc/roi_pool.h
index f528ce6d7e0..ac005914107 100644
--- a/torchvision/csrc/roi_pool.h
+++ b/torchvision/csrc/roi_pool.h
@@ -9,6 +9,9 @@
 #include "hip/roi_pool_kernel.h"
 #endif
 
+namespace vision {
+namespace ops {
+
 // C++ Forward
 std::tuple<at::Tensor, at::Tensor> roi_pool(
     const at::Tensor& input,
@@ -59,3 +62,6 @@ at::Tensor roi_pool_backward_autograd(
     int64_t channels,
     int64_t height,
     int64_t width);
+
+} // namespace ops
+} // namespace vision
diff --git a/torchvision/csrc/vision.cpp b/torchvision/csrc/vision.cpp
index fb0bf014912..b24b6c3df45 100644
--- a/torchvision/csrc/vision.cpp
+++ b/torchvision/csrc/vision.cpp
@@ -1,3 +1,5 @@
+#include "vision.h"
+
 #include <Python.h>
 #include <torch/script.h>
 
@@ -35,6 +37,8 @@ int64_t cuda_version() noexcept {
 }
 } // namespace vision
 
+using namespace vision::ops;
+
 TORCH_LIBRARY(torchvision, m) {
   m.def(
       "deform_conv2d(Tensor input, Tensor weight, Tensor offset, Tensor mask, Tensor bias, int stride_h, int stride_w, int pad_h, int pad_w, int dilation_h, int dilation_w, int groups, int offset_groups, bool use_mask) -> Tensor");
diff --git a/torchvision/csrc/vision.h b/torchvision/csrc/vision.h
index c99b25c030d..91935be6fdd 100644
--- a/torchvision/csrc/vision.h
+++ b/torchvision/csrc/vision.h
@@ -1,6 +1,5 @@
 #pragma once
 
-#include <torchvision/models/models.h>
 #include <cstdint>
 #include "macros.h"