Wrap FindKarcherMean for Rot3 #1069

johnwlambert · 2022-01-26T17:35:59Z

Wrap gtsam.FindKarcherMean with Rot3Vector argument (a new type defined in Rot3.h)

Previously, only KarcherMeanFactor was wrapped, meaning additional boilerplate (create a factor graph, create an optimizer, and optimize it) was required to compute the Karcher Mean.

Rot3Vector is a new type that is essentially a vector of Rot3 objects, although an Eigen aligned allocator is needed.

johnwlambert · 2022-01-26T18:56:15Z

gtsam/slam/slam.i

@@ -323,6 +323,9 @@ virtual class KarcherMeanFactor : gtsam::NonlinearFactor {
  KarcherMeanFactor(const gtsam::KeyVector& keys);
 };

+template <T = {gtsam::Rot3}>
+const T FindKarcherMean<T>(const std::vector<T>& rotations);


@varunagrawal @ProfFan question for you all -- do we have support for std::vector yet? or do we need to make a new type alias here such as Rot3Vector?

No because of MATLAB end

huh? I thought that std::vector worked with wrapper??

gtsam/slam/slam.i

akshay-krishnan · 2022-01-27T01:16:06Z

gtsam/slam/slam.i

+  gtsam::Rot3 at(size_t i) const;
+  void push_back(const gtsam::Rot3& R);
+};
+const gtsam::Rot3 FindKarcherMean<gtsam::Rot3>(const Rot3Vector& rotations);


no need for const in the return type: gtsam::Rot3 FindKarcherMean<gtsam::Rot3>(const Rot3Vector& rotations);

ProfFan · 2022-01-27T01:29:36Z

python/gtsam/specializations/slam.h

@@ -17,3 +17,4 @@ py::bind_vector<
 py::bind_vector<
    std::vector<boost::shared_ptr<gtsam::BetweenFactor<gtsam::Pose2> > > >(
    m_, "BetweenFactorPose2s");
+py::bind_vector<std::vector<gtsam::Rot3>>(m_, "Rot3Vector");


Are you sure you don't need aligned allocator?

Thanks for the review @ProfFan and @akshay-krishnan.

Let me check about aligned allocator. Btw I don't love the name Rot3Vector so much ... sounds like a vector representation of a Rot3. But Rot3Set would not be consistent with Pose3Vector nomenclature. So I guess it works.

ProfFan

LGTM minus one

johnwlambert · 2022-01-27T15:38:12Z

gtsam/slam/slam.i

+  void push_back(const gtsam::Rot3& R);
+};
+typedef FindKarcherMean<gtsam::Rot3> FindKarcherMeanRot3;
+gtsam::Rot3 FindKarcherMeanRot3(const Rot3Vector& rotations);


@dellaert @ProfFan the CI states: Cannot find class FindKarcherMean in module!. Strange. Should we be including #include <gtsam/slam/KarcherMeanFactor.h> instead of #include <gtsam/slam/KarcherMeanFactor-inl.h> in this .i file?

Traceback (most recent call last): File "/home/runner/work/gtsam/gtsam/wrap/scripts/pybind_wrap.py", line 94, in <module> main() File "/home/runner/work/gtsam/gtsam/wrap/scripts/pybind_wrap.py", line 85, in main wrapper.wrap_submodule(args.src) File "/home/runner/work/gtsam/gtsam/wrap/gtwrap/pybind_wrapper.py", line 696, in wrap_submodule cc_content = self.wrap_file(content, module_name=module_name) File "/home/runner/work/gtsam/gtsam/wrap/gtwrap/pybind_wrapper.py", line 626, in wrap_file module = instantiator.instantiate_namespace(module) File "/home/runner/work/gtsam/gtsam/wrap/gtwrap/template_instantiator/namespace.py", line 80, in instantiate_namespace element = instantiate_namespace(element) File "/home/runner/work/gtsam/gtsam/wrap/gtwrap/template_instantiator/namespace.py", line 59, in instantiate_namespace typedef_inst.typename) File "/home/runner/work/gtsam/gtsam/wrap/gtwrap/interface_parser/namespace.py", line 109, in find_class_or_function typename.name)) ValueError: Cannot find class FindKarcherMean in module! make[2]: *** [python/slam.cpp] Error 1 python/CMakeFiles/pybind_wrap_gtsam.dir/build.make:164: recipe for target 'python/slam.cpp' failed CMakeFiles/Makefile2:6769: recipe for target 'python/CMakeFiles/pybind_wrap_gtsam.dir/all' failed make[1]: *** [python/CMakeFiles/pybind_wrap_gtsam.dir/all] Error 2 make[1]: *** Waiting for unfinished jobs....

Should be the simple .h file. Is the FindKarcherMean class defined in the .i file?

FindKarcherMean is not defined in the .i file yet. I should add that, no that I think about it... (btw it is a function, not a class)

For some reason the inl.h file is being included, instead of the .h file currently. Not sure why though

Actually, since we can use std:: in .i, then we can't define FindKarcherMean in the .i file, since it relies on std::

template <class T> T FindKarcherMean(const std::vector<T, Eigen::aligned_allocator<T>> &rotations); template <class T> T FindKarcherMean(std::initializer_list<T> &&rotations);

So I guess the typedef for FindKarcherMeanRot3 needs to go into the .h file directly?

But you can use std::. Who is telling you you can't ?

Matlab wrapper needs correct signature, unless @varunagrawal implemented something that handles this?

If I'm not mistaken, std should be useable. When in doubt, check out or write a test for it in the wrap project. :)

(I spent a whole month writing those tests folks, please make them worth my time 😅)

Matlab wrapper needs correct signature, unless @varunagrawal implemented something that handles this?

I don't know what that means. Example?

dellaert · 2022-01-27T18:30:32Z

gtsam/slam/slam.i

+template <class T>
+T FindKarcherMean(const std::vector<T, Eigen::aligned_allocator<T>> &rotations);
+
+template <class T> T FindKarcherMean(std::initializer_list<T> &&rotations);


Not sure what this would translate to in python.....

dellaert · 2022-01-27T20:50:10Z

@johnwlambert , are you planning on still submitting other commits? You should know that on every push the CI runs 34 checks. It is recommended you do everything locally, including make check and make python-install, and then push once you think everything was addressed.

dellaert · 2022-01-27T20:50:57Z

If you are planning another push, kindly cancel the CI jobs for this PR.

johnwlambert · 2022-01-27T20:56:29Z

If you are planning another push, kindly cancel the CI jobs for this PR.

I've gotten to like the Github.com interface a bit too much : - ) I've cancelled the nonrelevant CI jobs, thanks.

dellaert · 2022-01-27T21:04:18Z

Yeah, please don't develop with our CI as your compute :-) It clogs all other PRs, and I'm trying to build a wheel (which will unfortunately not include this PR).

akshay-krishnan · 2022-01-28T07:30:35Z

python/gtsam/specializations/slam.h

@@ -17,3 +17,4 @@ py::bind_vector<
 py::bind_vector<
    std::vector<boost::shared_ptr<gtsam::BetweenFactor<gtsam::Pose2> > > >(
    m_, "BetweenFactorPose2s");
+py::bind_vector<std::vector<gtsam::Rot3, Eigen::aligned_allocator<gtsam::Rot3>>(m_, "Rot3Vector");


there is a missing ">" at the end here, that should let you wrap Rot3Vector at least.

akshay-krishnan · 2022-01-28T07:35:09Z

It looks like we cannot use std::initializer_list or Eigen::aligned_allocator in slam.i.
So the only way I can think of is to have a typedef for std::vector<Rot3, Eigen::aligned_allocator> and define another function specifically for Rot3 (using the typedef as the argument).
This seems a little hacky.
Alternatively, we could just re-implement FindKarcherMean in python using the KarcherMeanFactor which has been wrapped. Its a really small function. @johnwlambert WDYT?

johnwlambert · 2022-01-28T14:50:33Z

Thanks for taking a look @akshay-krishnan. Yeah this seems like we should just use KarcherMeanFactor directly in Python w/ an optimizer of our choice.

dellaert · 2022-01-28T14:53:45Z

Nonono! Let's fix this. Let me take a quick look.

dellaert · 2022-01-28T14:54:38Z

Why, @johnwlambert , there is not even a unit test here I can start from :-)

dellaert · 2022-01-28T14:58:37Z

@johnwlambert what were you actually trying to accomplish with this PR? There is no PR comment. And, there is test_KarcherMeanFactor.py that just works, I think?

johnwlambert · 2022-01-28T15:01:18Z

Thanks Frank. Yeah we were just trying to avoid repeating that boilerplate from https://github.com/borglab/gtsam/blob/develop/python/gtsam/tests/test_KarcherMeanFactor.py#L24. But we can use it in Python, no problem. A simple unit test would look like

import gtsam
from gtsam import Rot3, Rot3Vector
from gtsam.utils.test_case import GtsamTestCase


class TestKarcherMean(GtsamTestCase):
    """Tests for DecisionTreeFactors."""

    def test_find_karcher_mean_identity() -> None:
        """Averaging 3 identity rotations should yield the identity."""
        a1Rb1 = Rot3()
        a2Rb2 = Rot3()
        a3Rb3 = Rot3()

        aRb_list = Rot3Vector([a1Rb1, a2Rb2, a3Rb3])
        aRb_expected = Rot3()

        aRb = gtsam.FindKarcherMeanRot3(aRb_list)
        self.assertTrue(aRb.equals(aRb_expected))

but there are already some implemented here, which seem fine.

dellaert · 2022-01-28T17:01:15Z

@johnwlambert I think I fixed it, could you please update the PR with a good description after looking at the file differences?

johnwlambert · 2022-01-28T23:23:34Z

@johnwlambert I think I fixed it, could you please update the PR with a good description after looking at the file differences?

Beautiful work Frank. I added a bit to the description you wrote. Thanks!

johnwlambert added 3 commits January 26, 2022 10:35

Update slam.i

41384d1

fix typo

8ba9ae5

fix typo

b7a502e

johnwlambert commented Jan 26, 2022

View reviewed changes

johnwlambert added 5 commits January 26, 2022 17:35

fix template bug

055e027

add new type for vector of Rot3's

20622c2

add Rot3Vector type to specializations

9e63911

add Rot3Vector to python/CMakeLists.txt

6678fe0

make vector of Rot3's opaque

15c3dd4

akshay-krishnan reviewed Jan 27, 2022

View reviewed changes

akshay-krishnan requested a review from ProfFan January 27, 2022 01:17

ProfFan reviewed Jan 27, 2022

View reviewed changes

ProfFan approved these changes Jan 27, 2022

View reviewed changes

johnwlambert added 2 commits January 27, 2022 00:48

Address Akshay comments

72e4a9a

fix templated function

2f3fe73

johnwlambert commented Jan 27, 2022

View reviewed changes

define FindKarcherMean in .i file

8a22ffa

dellaert reviewed Jan 27, 2022

View reviewed changes

johnwlambert added 3 commits January 27, 2022 15:35

use Eigen::aligned_allocator

e0b37ed

use Eigen::aligned_allocator

56312d9

use aligned allocator for Rot3Vector

d37881e

akshay-krishnan reviewed Jan 28, 2022

View reviewed changes

johnwlambert closed this Jan 28, 2022

dellaert reopened this Jan 28, 2022

dellaert added 2 commits January 28, 2022 10:59

Merge branch 'develop' into wrap-karcher-mean-rot3

01f3fe5

Fix wrapping

55d8399

dellaert merged commit 3ba6566 into develop Jan 28, 2022

dellaert deleted the wrap-karcher-mean-rot3 branch January 28, 2022 20:41

varunagrawal mentioned this pull request Jan 30, 2022

Support direct STL container wrapping in the Pybind11 wrapper #1076

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Wrap FindKarcherMean for Rot3 #1069

Wrap FindKarcherMean for Rot3 #1069

johnwlambert commented Jan 26, 2022 •

edited

Loading

johnwlambert Jan 26, 2022

ProfFan Jan 27, 2022

dellaert Jan 27, 2022

akshay-krishnan Jan 27, 2022

ProfFan Jan 27, 2022

johnwlambert Jan 27, 2022

ProfFan left a comment

johnwlambert Jan 27, 2022

varunagrawal Jan 27, 2022

johnwlambert Jan 27, 2022

johnwlambert Jan 27, 2022 •

edited

Loading

dellaert Jan 27, 2022

ProfFan Jan 27, 2022

varunagrawal Jan 27, 2022

dellaert Jan 27, 2022

dellaert Jan 27, 2022

dellaert commented Jan 27, 2022

dellaert commented Jan 27, 2022

johnwlambert commented Jan 27, 2022

dellaert commented Jan 27, 2022

akshay-krishnan Jan 28, 2022

akshay-krishnan commented Jan 28, 2022

johnwlambert commented Jan 28, 2022

dellaert commented Jan 28, 2022

dellaert commented Jan 28, 2022

dellaert commented Jan 28, 2022

johnwlambert commented Jan 28, 2022

dellaert commented Jan 28, 2022

johnwlambert commented Jan 28, 2022

Wrap FindKarcherMean for Rot3 #1069

Wrap FindKarcherMean for Rot3 #1069

Conversation

johnwlambert commented Jan 26, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ProfFan left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

johnwlambert Jan 27, 2022 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dellaert commented Jan 27, 2022

dellaert commented Jan 27, 2022

johnwlambert commented Jan 27, 2022

dellaert commented Jan 27, 2022

Choose a reason for hiding this comment

akshay-krishnan commented Jan 28, 2022

johnwlambert commented Jan 28, 2022

dellaert commented Jan 28, 2022

dellaert commented Jan 28, 2022

dellaert commented Jan 28, 2022

johnwlambert commented Jan 28, 2022

dellaert commented Jan 28, 2022

johnwlambert commented Jan 28, 2022

johnwlambert commented Jan 26, 2022 •

edited

Loading

johnwlambert Jan 27, 2022 •

edited

Loading