Summary:
This is quite a thin wrapper – not sure we need it. The motivation is that `Transform3d` is not as matrix-centric now, it can be converted to SE(3) logarithm equally easily.
It simplifies things like averaging cameras and getting axis-angle of camera rotation (previously, one would need to call `se3_log_map(cameras.get_world_to_camera_transform().get_matrix())`), now one fewer thing to call / discover.
Reviewed By: bottler
Differential Revision: D39928000
fbshipit-source-id: 85248d5b8af136618f1d08791af5297ea5179d19
Summary:
One of the docstrings is a disaster see https://pytorch3d.readthedocs.io/en/latest/modules/ops.html
Also some minor fixes I encountered when browsing the code
Reviewed By: bottler
Differential Revision: D38581595
fbshipit-source-id: 3b6ca97788af380a44df9144a6a4cac782c6eab8
Summary:
Applies the black-fbsource codemod with the new build of pyfmt.
paintitblack
Reviewed By: lisroach
Differential Revision: D36324783
fbshipit-source-id: 280c09e88257e5e569ab729691165d8dedd767bc
Summary:
Previously, dtypes were not propagated correctly in composed transforms, resulting in errors when different dtypes were mixed. Even specifying a dtype in the constructor does not fix this. Neither does specifying the dtype for each composition function invocation (e.g. as a `kwarg` in `rotate_axis_angle`).
With the change, I also had to modify the default dtype of `RotateAxisAngle`, which was `torch.float64`; it is now `torch.float32` like for all other transforms. This was required because the fix in propagation broke some tests due to dtype mismatches.
This change in default dtype in turn broke two tests due to precision changes (calculations that were previously done in `torch.float64` were now done in `torch.float32`), so I changed the precision tolerances to be less strict. I chose the lowest power of ten that passed the tests here.
Pull Request resolved: https://github.com/facebookresearch/pytorch3d/pull/1141
Reviewed By: patricklabatut
Differential Revision: D35192970
Pulled By: bottler
fbshipit-source-id: ba0293e8b3595dfc94b3cf8048e50b7a5e5ed7cf
Summary: There are cases where importing pytorch3d seems to fail (internally at Meta) because of a clash between the builtin types module and ours, so rename ours.
Reviewed By: patricklabatut
Differential Revision: D34426817
fbshipit-source-id: f175448db6a4967a9a3f7bb6f595aad2ffb36455
Summary:
Add a test for Transform3d.stack, and make it work with composed transformations.
Fixes https://github.com/facebookresearch/pytorch3d/issues/1072 .
Reviewed By: patricklabatut
Differential Revision: D34211920
fbshipit-source-id: bfbd0895494ca2ad3d08a61bc82ba23637e168cc
Summary: Update all FB license strings to the new format.
Reviewed By: patricklabatut
Differential Revision: D33403538
fbshipit-source-id: 97a4596c5c888f3c54f44456dc07e718a387a02c
Summary:
Make sure the functions from `rotation_conversion` are jittable, and add some type hints.
Add tests to verify this is the case.
Pull Request resolved: https://github.com/facebookresearch/pytorch3d/pull/898
Reviewed By: patricklabatut
Differential Revision: D31926103
Pulled By: bottler
fbshipit-source-id: bff6013c5ca2d452e37e631bd902f0674d5ca091
Summary:
Copy some descriptions of renderer parameters to more places so they are easier to find.
Also a couple of small corrections, and make RasterizationSettings a dataclass.
Reviewed By: nikhilaravi, patricklabatut
Differential Revision: D30899822
fbshipit-source-id: 805cf366acb7d51cb308fa574deff0657c199673
Summary:
* HAT_INV_SKEW_SYMMETRIC_TOL was a global variable and torch script gives an error when compiling that function. Move it to the function scope.
* torch script gives error when compiling acos_linear_extrapolation because bound is a union of tuple and float. The tuple version is kept in this diff.
Reviewed By: patricklabatut
Differential Revision: D30614916
fbshipit-source-id: 34258d200dc6a09fbf8917cac84ba8a269c00aef
Summary: Restore compatibility with PyTorch 1.4 and 1.5, and a few lint fixes.
Reviewed By: patricklabatut
Differential Revision: D30048115
fbshipit-source-id: ee05efa7c625f6079fb06a3cc23be93e48df9433
Summary:
# Background
There is an unstable error during training (it can happen after several minutes or after several hours).
The error is connected to `torch.det()` function in `_check_valid_rotation_matrix()`.
if I remove the function `torch.det()` in `_check_valid_rotation_matrix()` or remove the whole functions `_check_valid_rotation_matrix()` the error is disappeared (D29555876).
# Solution
Replace `torch.det()` with manual implementation for 3x3 matrix.
Reviewed By: patricklabatut
Differential Revision: D29655924
fbshipit-source-id: 41bde1119274a705ab849751ece28873d2c45155
Summary:
# Make `transform3d.py` a little bit better (performance and code quality)
## 1. Add decorator `torch.no_grad()` to the function `_check_valid_rotation_matrix()`
Function `_check_valid_rotation_matrix()` is needed to identify errors during forward pass only, it's not used for gradients.
## 2. Replace two calls `to` with the single one
Reviewed By: bottler
Differential Revision: D29656501
fbshipit-source-id: 4419e24dbf436c1b60abf77bda4376fb87a593be
Summary: solve and lstsq have moved around in torch. Cope with both.
Reviewed By: patricklabatut
Differential Revision: D29302316
fbshipit-source-id: b34f0b923e90a357f20df359635929241eba6e74
Summary: Deprecate the `so3_exponential_map()` function in favor of its alias `so3_exp_map()`: this aligns with the naming of `so3_log_map()` and the recently introduced `se3_exp_map()` / `se3_log_map()` pair.
Reviewed By: bottler
Differential Revision: D29329966
fbshipit-source-id: b6f60b9e86b2995f70b1fbeb16f9feea05c55de9
Summary: Annotate the (return type of the) following dunder functions across the codebase: `__init__()`, `__len__()`, `__getitem__()`
Reviewed By: nikhilaravi
Differential Revision: D29001801
fbshipit-source-id: 928d9e1c417ffe01ab8c0445311287786e997c7c
Summary: Because rotations and (rotation) quaternions live on curved manifolds, it doesn't make sense to optimize them directly. Having a prominent option to require gradient on random ones may cause people to try, and isn't particularly useful.
Reviewed By: theschnitz
Differential Revision: D29160734
fbshipit-source-id: fc9e320672349fe334747c5b214655882a460a62
Summary:
Improves so3 functions to make gradient computation stable:
- Instead of `torch.acos`, uses `acos_linear_extrapolation` which has finite gradients of reasonable magnitude for all inputs.
- Adds tests for the latter.
The tests of the finiteness of the gradient in `test_so3_exp_singularity`, `test_so3_exp_singularity`, `test_so3_cos_bound` would fail if the `so3` functions would call `torch.acos` instead of `acos_linear_extrapolation`.
Reviewed By: bottler
Differential Revision: D23326429
fbshipit-source-id: dc296abf2ae3ddfb3942c8146621491a9cb740ee
Summary:
Implements a backprop-safe version of `torch.acos` that linearly extrapolates the function outside bounds.
Below is a plot of the extrapolated acos for different bounds:
{F611339485}
Reviewed By: bottler, nikhilaravi
Differential Revision: D27945714
fbshipit-source-id: fa2e2385b56d6fe534338d5192447c4a3aec540c
Summary:
As reported on github, `matrix_to_quaternion` was incorrect for rotations by 180˚. We resolved the sign of the component `i` based on the sign of `i*r`, assuming `r > 0`, which is untrue if `r == 0`.
This diff handles special cases and ensures we use the non-zero elements to copy the sign from.
Reviewed By: bottler
Differential Revision: D29149465
fbshipit-source-id: cd508cc31567fc37ea3463dd7e8c8e8d5d64a235
Summary: Make Transform3d.to() not ignore a different dtype when device is the same and no copy is requested. Fix other methods where dtype is ignored.
Reviewed By: nikhilaravi
Differential Revision: D28981171
fbshipit-source-id: 4528e6092f4a693aecbe8131ede985fca84e84cf
Summary:
Tidy uses of `torch.device` in `Transforms3d`:
- Allow `str` or `torch.device` in user-facing methods
- Consistently use `torch.device` for internal types
- Fix comparison of devices
Reviewed By: nikhilaravi
Differential Revision: D28929486
fbshipit-source-id: bd1d6cc7ede3d8fd549fd3224a9b07eec53f8164
Summary: Currently the Rotate transform does not consider the R's device at all, resulting in errors if you're expecting it to be on cuda but it gets the default casting to cpu. This updates the transform to respect R's device.
Reviewed By: nikhilaravi
Differential Revision: D27828118
fbshipit-source-id: ddd99f73eadbd990688eb22f3d1ffbacbe168c81
Summary: Make `to` on Transform3D carry its member _transforms.
Reviewed By: nikhilaravi
Differential Revision: D25978611
fbshipit-source-id: 12b39e7a657f28d59ca60800bf9f4193a2c08197
Summary:
Defines a function to run marching cubes algorithm on a single or batch of 3D scalar fields. Returns a mesh's faces and vertices.
UPDATES (12/18)
- Input data is now specified as a (B, D, H, W) tensor as opposed to a (B, W, H, D) tensor. This will now be compatible with the Volumes datastructure.
- Add an option to return output vertices in local coordinates instead of world coordinates.
Also added a small fix to remove the dype for device in Transforms3D - if passing in a torch.device instead of str it causes a pyre error.
Reviewed By: jcjohnson
Differential Revision: D24599019
fbshipit-source-id: 90554a200319fed8736a12371cc349e7108aacd0
Summary:
This diff builds on top of the `pulsar integration` diff to provide a unified interface for the existing PyTorch3D point renderer and Pulsar. For more information about the pulsar backend, see the release notes and the paper (https://arxiv.org/abs/2004.07484). For information on how to use the backend, see the point cloud rendering notebook and the examples in the folder docs/examples.
The unified interfaces are completely consistent. Switching the render backend is as easy as using `renderer = PulsarPointsRenderer(rasterizer=rasterizer).to(device)` instead of `renderer = PointsRenderer(rasterizer=rasterizer, compositor=compositor)` and adding the `gamma` parameter to the forward function. All PyTorch3D camera types are supported as far as possible; keyword arguments are properly forwarded to the camera. The `PerspectiveCamera` and `OrthographicCamera` require znear and zfar as additional parameters for the forward pass.
Reviewed By: nikhilaravi
Differential Revision: D21421443
fbshipit-source-id: 4aa0a83a419592d9a0bb5d62486a1cdea9d73ce6
Summary:
This diff integrates the pulsar renderer source code into PyTorch3D as an alternative backend for the PyTorch3D point renderer. This diff is the first of a series of three diffs to complete that migration and focuses on the packaging and integration of the source code.
For more information about the pulsar backend, see the release notes and the paper (https://arxiv.org/abs/2004.07484). For information on how to use the backend, see the point cloud rendering notebook and the examples in the folder `docs/examples`.
Tasks addressed in the following diffs:
* Add the PyTorch3D interface,
* Add notebook examples and documentation (or adapt the existing ones to feature both interfaces).
Reviewed By: nikhilaravi
Differential Revision: D23947736
fbshipit-source-id: a5e77b53e6750334db22aefa89b4c079cda1b443
Summary: Fix axis_angle conversions where I used torch.square which doesn't work with pytorch 1.4
Reviewed By: nikhilaravi
Differential Revision: D24451546
fbshipit-source-id: ba26f7dad5fa991f0a8f7d3d09ee7151163aecf4
Summary: We can represent a rotation as a vector in the axis direction, whose length is the rotation anticlockwise in radians around that axis.
Reviewed By: gkioxari
Differential Revision: D24306293
fbshipit-source-id: 2e0f138eda8329f6cceff600a6e5f17a00e4deb7
Summary: Issue #119. The function `sqrt(max(x, 0))` is not convex and has infinite gradient at 0, but 0 is a subgradient at 0. Here we implement it in such a way as to give 0 as the gradient.
Reviewed By: gkioxari
Differential Revision: D24306294
fbshipit-source-id: 48d136faca083babad4d64970be7ea522dbe9e09
