Summary:
Update the cuda kernels to:
- remove contiguous checks for the grad tensors and for cpu functions which use accessors
- for cuda implementations call `.contiguous()` on all tensors in the host function before invoking the kernel
Reviewed By: gkioxari
Differential Revision: D21598008
fbshipit-source-id: 9b97bda4582fd4269c8a00999874d4552a1aea2d
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
Summary: This has been failing intermittently
Reviewed By: nikhilaravi
Differential Revision: D21403157
fbshipit-source-id: 51b74d6c813b52effe72d14b565e250fcabbb463
Summary:
Updates to:
- enable cuda kernel launches on any GPU (not just the default)
- cuda and contiguous checks for all kernels
- checks to ensure all tensors are on the same device
- error reporting in the cuda kernels
- cuda tests now run on a random device not just the default
Reviewed By: jcjohnson, gkioxari
Differential Revision: D21215280
fbshipit-source-id: 1bedc9fe6c35e9e920bdc4d78ed12865b1005519
Summary:
Use aten instead of torch interface in all cuda code. This allows the cuda build to work with pytorch 1.5 with GCC 5 (e.g. the compiler of ubuntu 16.04LTS). This wasn't working. It has been failing with errors like the below, perhaps due to a bug in nvcc.
```
torch/include/torch/csrc/api/include/torch/nn/cloneable.h:68:61: error: invalid static_cast from type ‘const torch::OrderedDict<std::basic_string<char>, std::shared_ptr<torch::nn::Module> >’ to type ‘torch::OrderedDict<std::basic_string<char>, std::shared_ptr<torch::nn::Module> >
```
Reviewed By: nikhilaravi
Differential Revision: D21204029
fbshipit-source-id: ca6bdbcecf42493365e1c23a33fe35e1759fe8b6
Summary:
We have multiple KNN CUDA implementations. From python, users can currently request a particular implementation via the `version` flag, but they have no way of knowing which implementations can be used for a given problem.
This diff exposes a function `pytorch3d._C.knn_check_version(version, D, K)` that returns whether a particular version can be used.
Reviewed By: nikhilaravi
Differential Revision: D21162573
fbshipit-source-id: 6061960bdcecba454fd920b00036f4e9ff3fdbc0
Summary:
Added backface culling as an option to the `raster_settings`. This is needed for the full forward rendering of shapenet meshes with texture (some meshes contain
multiple overlapping segments which have different textures).
For a triangle (v0, v1, v2) define the vectors A = (v1 - v0) and B = (v2 − v0) and use this to calculate the area of the triangle as:
```
area = 0.5 * A x B
area = 0.5 * ((x1 − x0)(y2 − y0) − (x2 − x0)(y1 − y0))
```
The area will be positive if (v0, v1, v2) are oriented counterclockwise (a front face), and negative if (v0, v1, v2) are oriented clockwise (a back face).
We can reuse the `edge_function` as it already calculates the triangle area.
Reviewed By: jcjohnson
Differential Revision: D20960115
fbshipit-source-id: 2d8a4b9ccfb653df18e79aed8d05c7ec0f057ab1
Summary:
Fix a bug which resulted in a rendering artifacts if the image size was not a multiple of 16.
Fix: Revert coarse rasterization to original implementation and only update fine rasterization to reverse the ordering of Y and X axis. This is much simpler than the previous approach!
Additional changes:
- updated mesh rendering end-end tests to check outputs from both naive and coarse to fine rasterization.
- added pointcloud rendering end-end tests
Reviewed By: gkioxari
Differential Revision: D21102725
fbshipit-source-id: 2e7e1b013dd6dd12b3a00b79eb8167deddb2e89a
Summary: This is mostly replacing the old PackedTensorAccessor with the new PackedTensorAccessor64.
Reviewed By: gkioxari
Differential Revision: D21088773
fbshipit-source-id: 5973e5a29d934eafb7c70ec5ec154ca076b64d27
Summary: A couple of files for the removed nearest_neighbor functionality are left behind.
Reviewed By: nikhilaravi
Differential Revision: D21088624
fbshipit-source-id: 4bb29016b4e5f63102765b384c363733b60032fa
Summary: knn is more general and faster than the nearest_neighbor code, so remove the latter.
Reviewed By: gkioxari
Differential Revision: D20816424
fbshipit-source-id: 75d6c44d17180752d0c9859814bbdf7892558158
Summary: Interface and working implementation of ragged KNN. Benchmarks (which aren't ragged) haven't slowed. New benchmark shows that ragged is faster than non-ragged of the same shape.
Reviewed By: jcjohnson
Differential Revision: D20696507
fbshipit-source-id: 21b80f71343a3475c8d3ee0ce2680f92f0fae4de
Summary: Run linter after recent changes. Fix long comment in knn.h which clang-format has reflowed badly. Add crude test that code doesn't call deprecated `.type()` or `.data()`.
Reviewed By: nikhilaravi
Differential Revision: D20692935
fbshipit-source-id: 28ce0308adae79a870cb41a810b7cf8744f41ab8
Summary:
Implements K-Nearest Neighbors with C++ and CUDA versions.
KNN in CUDA is highly nontrivial. I've implemented a few different versions of the kernel, and we heuristically dispatch to different kernels based on the problem size. Some of the kernels rely on template specialization on either D or K, so we use template metaprogramming to compile specialized versions for ranges of D and K.
These kernels are up to 3x faster than our existing 1-nearest-neighbor kernels, so we should also consider swapping out `nn_points_idx` to use these kernels in the backend.
I've been working mostly on the CUDA kernels, and haven't converged on the correct Python API.
I still want to benchmark against FAISS to see how far away we are from their performance.
Reviewed By: bottler
Differential Revision: D19729286
fbshipit-source-id: 608ffbb7030c21fe4008f330522f4890f0c3c21a
Summary: CPU-only builds should be fixed by this change
Reviewed By: nikhilaravi
Differential Revision: D20598014
fbshipit-source-id: df098ec4c6c93d38515172805fe57cac7463c506
Summary: Comments were describing squared distance as absolute distance in a few places.
Reviewed By: nikhilaravi
Differential Revision: D20426020
fbshipit-source-id: 009946867c4a98f61f5ce7158542d41e22bf8346
Summary:
Applying the changes added for mesh rasterization to ensure that +Y is up and +X is left so that the coordinate system is right handed.
Also updated the diagram in the docs to indicate that (0,0) is in the top left hand corner.
Reviewed By: gkioxari
Differential Revision: D20394849
fbshipit-source-id: cfb7c79090eb1f55ad38b92327a74a70a8dc541e
Summary:
Update the point cloud rasterizer to:
- use the pointcloud datastructure (rebased on top of D19791851.)
- support rasterization of heterogeneous point clouds in the same way as with Meshes.
The main changes to the API will be as follows:
- The input to `rasterize_points` will be a `Pointclouds` object instead of a tensor. This will be easy to update e.g.
```
points = torch.randn(N, P, 3)
idx2, zbuf2, dists2 = rasterize_points(points, image_size, radius, points_per_pixel)
points = torch.randn(N, P, 3)
pointclouds = Pointclouds(points=points)
idx2, zbuf2, dists2 = rasterize_points(pointclouds, image_size, radius, points_per_pixel)
```
- The indices output from rasterization will now refer to points in `poinclouds.points_packed()`.
This may require some changes to the functions which consume the outputs of rasterization if they were previously
assuming that the indices ranged from 0 to P where P is the number of points in each pointcloud.
Making this change now so that Olivia can update her PR accordingly.
Reviewed By: gkioxari
Differential Revision: D20088651
fbshipit-source-id: 833ed659909712bcbbb6a50e2ec0189839f0413a
Summary:
## Updates
- Defined the world and camera coordinates according to this figure. The world coordinates are defined as having +Y up, +X left and +Z in.
{F230888499}
- Removed all flipping from blending functions.
- Updated the rasterizer to return images with +Y up and +X left.
- Updated all the mesh rasterizer tests
- The expected values are now defined in terms of the default +Y up, +X left
- Added tests where the triangles in the meshes are non symmetrical so that it is clear which direction +X and +Y are
## Questions:
- Should we have **scene settings** instead of raster settings?
- To be more correct we should be [z clipping in the rasterizer based on the far/near clipping planes](https://github.com/ShichenLiu/SoftRas/blob/master/soft_renderer/cuda/soft_rasterize_cuda_kernel.cu#L400) - these values are also required in the blending functions so should we make these scene level parameters and have a scene settings tuple which is available to the rasterizer and shader?
Reviewed By: gkioxari
Differential Revision: D20208604
fbshipit-source-id: 55787301b1bffa0afa9618f0a0886cc681da51f3
Summary:
`PointLineDistanceForward()` should return squared distance. However, it seems that it returned non-squared distance when `v0` was near by `v1` in CPU implementation.
Pull Request resolved: https://github.com/facebookresearch/pytorch3d/pull/83
Reviewed By: bottler
Differential Revision: D20097181
Pulled By: nikhilaravi
fbshipit-source-id: 7ea851c0837ab89364e42d283c999df21ff5ff02
Summary:
Fixed a few MSVC compiler (visual studio 2019, MSVC 19.16.27034) compatibility issues
1. Replaced long with int64_t. aten::data_ptr\<long\> is not supported in MSVC
2. pytorch3d/csrc/rasterize_points/rasterize_points_cpu.cpp, inline function is not correctly recognized by MSVC.
3. pytorch3d/csrc/rasterize_meshes/geometry_utils.cuh
const auto kEpsilon = 1e-30;
MSVC does not compile this const into both host and device, change to a MACRO.
4. pytorch3d/csrc/rasterize_meshes/geometry_utils.cuh,
const float area2 = pow(area, 2.0);
2.0 is considered as double by MSVC and raised an error
5. pytorch3d/csrc/rasterize_points/rasterize_points_cpu.cpp
std::tuple<torch::Tensor, torch::Tensor> RasterizePointsCoarseCpu() return type does not match the declaration in rasterize_points_cpu.h.
Pull Request resolved: https://github.com/facebookresearch/pytorch3d/pull/9
Reviewed By: nikhilaravi
Differential Revision: D19986567
Pulled By: yuanluxu
fbshipit-source-id: f4d98525d088c99c513b85193db6f0fc69c7f017
Summary:
Added backward for mesh face areas & normals. Exposed it as a layer. Replaced the computation with the new op in Meshes and in Sample Points.
Current issue: Circular imports. I moved the import of the op in meshes inside the function scope.
Reviewed By: jcjohnson
Differential Revision: D19920082
fbshipit-source-id: d213226d5e1d19a0c8452f4d32771d07e8b91c0a
Summary:
pybind now seems to need C++17 on a mac, so advise people to use it. (Also delete an unused variable to silence a warning I got on a mac build.)
Reported in github issue #68.
Reviewed By: nikhilaravi
Differential Revision: D19970512
fbshipit-source-id: f9be20c8ed425bd6ba8d009a7d62dad658dccdb1
Summary:
Added if `WITH_CUDA` checks for points/mesh rasterization. If installing on cpu only then this causes `Undefined symbol` errors when trying to import pytorch3d.
We had these checks for all the other cuda files but not the rasterization files.
Thanks ppwwyyxx for the tip!
Reviewed By: ppwwyyxx, gkioxari
Differential Revision: D19801495
fbshipit-source-id: 20e7adccfdb33ac731c00a89414b2beaf0a35529
Summary:
Adds a CPU implementation for `pytorch3d.ops.nn_points_idx`.
Also renames the associated C++ and CUDA functions to use `AllCaps` names used in other C++ / CUDA code.
Reviewed By: gkioxari
Differential Revision: D19670491
fbshipit-source-id: 1b6409404025bf05e6a93f5d847e35afc9062f05