make C extension lazy-import (#971)

* make C extension lazy-import

* add lazy loading to roi_pool

torchvision/__init__.py

@@ -33,31 +33,3 @@ def get_image_backend():
     Gets the name of the package used to load images
     """
     return _image_backend
-
-
-def _check_cuda_matches():
-    """
-    Make sure that CUDA versions match between the pytorch install and torchvision install
-    """
-    import torch
-    from torchvision import _C
-    if hasattr(_C, "CUDA_VERSION") and torch.version.cuda is not None:
-        tv_version = str(_C.CUDA_VERSION)
-        if int(tv_version) < 10000:
-            tv_major = int(tv_version[0])
-            tv_minor = int(tv_version[2])
-        else:
-            tv_major = int(tv_version[0:2])
-            tv_minor = int(tv_version[3])
-        t_version = torch.version.cuda
-        t_version = t_version.split('.')
-        t_major = int(t_version[0])
-        t_minor = int(t_version[1])
-        if t_major != tv_major or t_minor != tv_minor:
-            raise RuntimeError("Detected that PyTorch and torchvision were compiled with different CUDA versions. "
-                               "PyTorch has CUDA Version={}.{} and torchvision has CUDA Version={}.{}. "
-                               "Please reinstall the torchvision that matches your PyTorch install."
-                               .format(t_major, t_minor, tv_major, tv_minor))
-
-
-_check_cuda_matches()

torchvision/extension.py

+_C = None
+
+
+def _lazy_import():
+    """
+    Make sure that CUDA versions match between the pytorch install and torchvision install
+    """
+    global _C
+    if _C is not None:
+        return _C
+    import torch
+    from torchvision import _C as C
+    _C = C
+    if hasattr(_C, "CUDA_VERSION") and torch.version.cuda is not None:
+        tv_version = str(_C.CUDA_VERSION)
+        if int(tv_version) < 10000:
+            tv_major = int(tv_version[0])
+            tv_minor = int(tv_version[2])
+        else:
+            tv_major = int(tv_version[0:2])
+            tv_minor = int(tv_version[3])
+        t_version = torch.version.cuda
+        t_version = t_version.split('.')
+        t_major = int(t_version[0])
+        t_minor = int(t_version[1])
+        if t_major != tv_major or t_minor != tv_minor:
+            raise RuntimeError("Detected that PyTorch and torchvision were compiled with different CUDA versions. "
+                               "PyTorch has CUDA Version={}.{} and torchvision has CUDA Version={}.{}. "
+                               "Please reinstall the torchvision that matches your PyTorch install."
+                               .format(t_major, t_minor, tv_major, tv_minor))
+    return _C

torchvision/ops/boxes.py

 import torch
-from torchvision import _C
+from torchvision.extension import _lazy_import
 
 
 def nms(boxes, scores, iou_threshold):
@@ -22,6 +22,7 @@ def nms(boxes, scores, iou_threshold):
             of the elements that have been kept
             by NMS, sorted in decreasing order of scores
     """
+    _C = _lazy_import()
     return _C.nms(boxes, scores, iou_threshold)
 
 

torchvision/ops/roi_align.py

@@ -6,7 +6,7 @@ from torch.autograd.function import once_differentiable
 
 from torch.nn.modules.utils import _pair
 
-from torchvision import _C
+from torchvision.extension import _lazy_import
 from ._utils import convert_boxes_to_roi_format
 
 
@@ -18,6 +18,7 @@ class _RoIAlignFunction(Function):
         ctx.spatial_scale = spatial_scale
         ctx.sampling_ratio = sampling_ratio
         ctx.input_shape = input.size()
+        _C = _lazy_import()
         output = _C.roi_align_forward(
             input, roi, spatial_scale,
             output_size[0], output_size[1], sampling_ratio)
@@ -31,6 +32,7 @@ class _RoIAlignFunction(Function):
         spatial_scale = ctx.spatial_scale
         sampling_ratio = ctx.sampling_ratio
         bs, ch, h, w = ctx.input_shape
+        _C = _lazy_import()
         grad_input = _C.roi_align_backward(
             grad_output, rois, spatial_scale,
             output_size[0], output_size[1], bs, ch, h, w, sampling_ratio)

torchvision/ops/roi_pool.py

@@ -6,7 +6,7 @@ from torch.autograd.function import once_differentiable
 
 from torch.nn.modules.utils import _pair
 
-from torchvision import _C
+from torchvision.extension import _lazy_import
 from ._utils import convert_boxes_to_roi_format
 
 
@@ -16,6 +16,7 @@ class _RoIPoolFunction(Function):
         ctx.output_size = _pair(output_size)
         ctx.spatial_scale = spatial_scale
         ctx.input_shape = input.size()
+        _C = _lazy_import()
         output, argmax = _C.roi_pool_forward(
             input, rois, spatial_scale,
             output_size[0], output_size[1])
@@ -29,6 +30,7 @@ class _RoIPoolFunction(Function):
         output_size = ctx.output_size
         spatial_scale = ctx.spatial_scale
         bs, ch, h, w = ctx.input_shape
+        _C = _lazy_import()
         grad_input = _C.roi_pool_backward(
             grad_output, rois, argmax, spatial_scale,
             output_size[0], output_size[1], bs, ch, h, w)