avoid converting a TensorOptions from float to integer

Summary: pytorch is adding checks that mean integer tensors with requires_grad=True need to be avoided. Fix accidentally creating them.

Reviewed By: jcjohnson, gkioxari

Differential Revision: D21576712

fbshipit-source-id: 008218997986800a36d93caa1a032ee91f2bffcd

pytorch3d/csrc/knn/knn.cu

@@ -319,7 +319,7 @@ std::tuple<at::Tensor, at::Tensor> KNearestNeighborIdxCuda(
   const int64_t K_64 = K;
 
   TORCH_CHECK(p2.size(2) == D, "Point sets must have the same last dimension");
-  auto long_dtype = p1.options().dtype(at::kLong);
+  auto long_dtype = lengths1.options().dtype(at::kLong);
   auto idxs = at::zeros({N, P1, K}, long_dtype);
   auto dists = at::zeros({N, P1, K}, p1.options());
 

pytorch3d/csrc/knn/knn_cpu.cpp

@@ -14,7 +14,7 @@ std::tuple<at::Tensor, at::Tensor> KNearestNeighborIdxCpu(
   const int P1 = p1.size(1);
   const int D = p1.size(2);
 
-  auto long_opts = p1.options().dtype(torch::kInt64);
+  auto long_opts = lengths1.options().dtype(torch::kInt64);
   torch::Tensor idxs = torch::full({N, P1, K}, 0, long_opts);
   torch::Tensor dists = torch::full({N, P1, K}, 0, p1.options());
 

pytorch3d/csrc/rasterize_meshes/rasterize_meshes.cu

@@ -321,7 +321,7 @@ RasterizeMeshesNaiveCuda(
   const int W = image_size;
   const int K = num_closest;
 
-  auto long_opts = face_verts.options().dtype(at::kLong);
+  auto long_opts = num_faces_per_mesh.options().dtype(at::kLong);
   auto float_opts = face_verts.options().dtype(at::kFloat);
 
   at::Tensor face_idxs = at::full({N, H, W, K}, -1, long_opts);
@@ -701,7 +701,7 @@ at::Tensor RasterizeMeshesCoarseCuda(
     ss << "Got " << num_bins << "; that's too many!";
     AT_ERROR(ss.str());
   }
-  auto opts = face_verts.options().dtype(at::kInt);
+  auto opts = num_faces_per_mesh.options().dtype(at::kInt);
   at::Tensor faces_per_bin = at::zeros({N, num_bins, num_bins}, opts);
   at::Tensor bin_faces = at::full({N, num_bins, num_bins, M}, -1, opts);
 
@@ -868,7 +868,7 @@ RasterizeMeshesFineCuda(
   if (K > kMaxPointsPerPixel) {
     AT_ERROR("Must have num_closest <= 150");
   }
-  auto long_opts = face_verts.options().dtype(at::kLong);
+  auto long_opts = bin_faces.options().dtype(at::kLong);
   auto float_opts = face_verts.options().dtype(at::kFloat);
 
   at::Tensor face_idxs = at::full({N, H, W, K}, -1, long_opts);

pytorch3d/csrc/rasterize_meshes/rasterize_meshes_cpu.cpp

@@ -123,7 +123,7 @@ RasterizeMeshesNaiveCpu(
   const int W = image_size;
   const int K = faces_per_pixel;
 
-  auto long_opts = face_verts.options().dtype(torch::kInt64);
+  auto long_opts = num_faces_per_mesh.options().dtype(torch::kInt64);
   auto float_opts = face_verts.options().dtype(torch::kFloat32);
 
   // Initialize output tensors.
@@ -418,7 +418,7 @@ torch::Tensor RasterizeMeshesCoarseCpu(
   const int BH = 1 + (height - 1) / bin_size; // Integer division round up.
   const int BW = 1 + (width - 1) / bin_size; // Integer division round up.
 
-  auto opts = face_verts.options().dtype(torch::kInt32);
+  auto opts = num_faces_per_mesh.options().dtype(torch::kInt32);
   torch::Tensor faces_per_bin = torch::zeros({N, BH, BW}, opts);
   torch::Tensor bin_faces = torch::full({N, BH, BW, M}, -1, opts);
   auto bin_faces_a = bin_faces.accessor<int32_t, 4>();

pytorch3d/csrc/rasterize_points/rasterize_points.cu

@@ -177,7 +177,7 @@ std::tuple<at::Tensor, at::Tensor, at::Tensor> RasterizePointsNaiveCuda(
     AT_ERROR(ss.str());
   }
 
-  auto int_opts = points.options().dtype(at::kInt);
+  auto int_opts = num_points_per_cloud.options().dtype(at::kInt);
   auto float_opts = points.options().dtype(at::kFloat);
   at::Tensor point_idxs = at::full({N, S, S, K}, -1, int_opts);
   at::Tensor zbuf = at::full({N, S, S, K}, -1, float_opts);
@@ -372,7 +372,7 @@ at::Tensor RasterizePointsCoarseCuda(
     ss << "Got " << num_bins << "; that's too many!";
     AT_ERROR(ss.str());
   }
-  auto opts = points.options().dtype(at::kInt);
+  auto opts = num_points_per_cloud.options().dtype(at::kInt);
   at::Tensor points_per_bin = at::zeros({N, num_bins, num_bins}, opts);
   at::Tensor bin_points = at::full({N, num_bins, num_bins, M}, -1, opts);
 
@@ -509,7 +509,7 @@ std::tuple<at::Tensor, at::Tensor, at::Tensor> RasterizePointsFineCuda(
   if (K > kMaxPointsPerPixel) {
     AT_ERROR("Must have num_closest <= 150");
   }
-  auto int_opts = points.options().dtype(at::kInt);
+  auto int_opts = bin_points.options().dtype(at::kInt);
   auto float_opts = points.options().dtype(at::kFloat);
   at::Tensor point_idxs = at::full({N, S, S, K}, -1, int_opts);
   at::Tensor zbuf = at::full({N, S, S, K}, -1, float_opts);

pytorch3d/csrc/rasterize_points/rasterize_points_cpu.cpp

@@ -25,7 +25,7 @@ std::tuple<torch::Tensor, torch::Tensor, torch::Tensor> RasterizePointsNaiveCpu(
   const int K = points_per_pixel;
 
   // Initialize output tensors.
-  auto int_opts = points.options().dtype(torch::kInt32);
+  auto int_opts = num_points_per_cloud.options().dtype(torch::kInt32);
   auto float_opts = points.options().dtype(torch::kFloat32);
   torch::Tensor point_idxs = torch::full({N, S, S, K}, -1, int_opts);
   torch::Tensor zbuf = torch::full({N, S, S, K}, -1, float_opts);
@@ -105,7 +105,7 @@ torch::Tensor RasterizePointsCoarseCpu(
 
   const int B = 1 + (image_size - 1) / bin_size; // Integer division round up
   const int M = max_points_per_bin;
-  auto opts = points.options().dtype(torch::kInt32);
+  auto opts = num_points_per_cloud.options().dtype(torch::kInt32);
   torch::Tensor points_per_bin = torch::zeros({N, B, B}, opts);
   torch::Tensor bin_points = torch::full({N, B, B, M}, -1, opts);
 

pytorch3d/ops/knn.py

@@ -67,6 +67,7 @@ class _knn_points(Function):
             idx = idx.gather(2, sort_idx)
 
         ctx.save_for_backward(p1, p2, lengths1, lengths2, idx)
+        ctx.mark_non_differentiable(idx)
         return dists, idx
 
     @staticmethod

pytorch3d/renderer/mesh/rasterize_meshes.py

@@ -197,6 +197,7 @@ class _RasterizeFaceVerts(torch.autograd.Function):
             cull_backfaces,
         )
         ctx.save_for_backward(face_verts, pix_to_face)
+        ctx.mark_non_differentiable(pix_to_face)
         ctx.perspective_correct = perspective_correct
         return pix_to_face, zbuf, barycentric_coords, dists
 

pytorch3d/renderer/points/rasterize_points.py

@@ -140,6 +140,7 @@ class _RasterizePoints(torch.autograd.Function):
         )
         idx, zbuf, dists = _C.rasterize_points(*args)
         ctx.save_for_backward(points, idx)
+        ctx.mark_non_differentiable(idx)
         return idx, zbuf, dists
 
     @staticmethod