darglein / snake-slam Goto Github PK

@darglein
Hi,thank you very much for your work, I was very interested in this, so went to cmake and compile the project, But encountered various problems.
My computer default environment is : ubuntu18.04,g++ version 7.5(default) cuda 10.2
In the beginning, cmake could not succeed in such an environment, so I upgraded g++ to 8.4 and cuda to 11.1, and cmake could succeed, but the following error occurred when compiling:
error 1:
/lib/libsaiga_core.so: undefined reference to 'vtable for std::filesystem::__cxx11::filesystem_error'
error 2:
Snake-SLAM/External/saiga/src/saiga/vision/imu/DecoupledImuSolverCeres.cpp:276:24: error: no member named 'evaluation_callback' in 'ceres::Problem::Options'

@darglein
Thank you very much for your work, I read the snake-slam paper and thought it was great, but when I compile the program, there are various errors, some internal errors, such as in SnakeGlobal.h:
using Vec4f = Saiga::Vec4f;

using Vec3f = Saiga::Vec3f;

using Vec2f = Saiga::Vec2f;
error: no type named 'Vec4f' in namespace 'Saiga'
Next:
inline Intrinsics4& K = mono_intrinsics.model.K;
error: unknown type name 'Intrinsics4'
Intrinsics4 is a file name, but it is used as a definition variable in the program, which is obviously wrong. I changed it to the following:
inline IntrinsicsPinholed& K = mono_intrinsics.model.K;
The program runs normally

In addition to these bugs, there are some bugs in FeatureDetector.h and FeatureDetector.cpp. So I would like to ask, is your program uploading the full version or the correct version?
Had very high expectations for the snake-slam framework because of your paper and seeing your framework in a SenseTime competition, but the bugs did disappoint me a bit，so looking forward to your early reply！

Hi I got the following error while compiling, kindly help me with that.

usr/bin/ld: cannot find -lomp

Hi, and thank you for making this code available. I would like to test running it live with the Stereolabs Zed 2 camera, would this be fairly straightforward to implement? Is there any reason why the code would not work with this input sensor? (Stereo frames + IMU).

Thanks!

@darglein
Hello, when I run the TUM RGBD data (rgbd_dataset_freiburg3_long_office_household), considering that the snake slam has the function of dense reconstruction, when I click on the fuse depth map, the following pipeline appears, but the saiga interface does not show dense, how to solve this problem or let which outputs the final result？

Initialization done.
clear map.
CudaStream::setName only working if you enable SAIGA_CUDA_WITH_NVTOOLS in cmake
CudaStream::setName only working if you enable SAIGA_CUDA_WITH_NVTOOLS in cmake
CudaStream::setName only working if you enable SAIGA_CUDA_WITH_NVTOOLS in cmake
CudaStream::setName only working if you enable SAIGA_CUDA_WITH_NVTOOLS in cmake
Assertion 'num_elements > 0' failed!
File: /mnt/Disk0/2_Linux/0_myStuff/Snake-SLAM/External/saiga/src/saiga/opengl/UnifiedMeshBuffer.cpp:105
Function: void Saiga::UnifiedMeshBuffer::Draw(int, int)
[1] 30871 abort (core dumped) ./build/bin/snake_slam configs/euroc.ini

i can't close saiga window and program abort.

how can i fix it without cuda ?

@darglein

When I run snake slam, saiga doesn't visualize the tracking result of the slam, saiga is black and nothing!!!!!!!!!

@darglein
Thank your great work and opensource code，but when I compile the code,there is some error blow:

/home/divenire/Public/Snake-SLAM/Snake/Preprocess/Input.cpp:163:46: error: no member named 'ZJUDataset' in namespace 'Saiga'
auto c = std::make_uniqueSaiga::ZJUDataset(settings.datasetParameters);
~~~~~~~^
/home/divenire/Public/Snake-SLAM/Snake/Preprocess/Input.cpp:176:46: error: no member named 'EuRoCDataset' in namespace 'Saiga'
auto c = std::make_uniqueSaiga::EuRoCDataset(settings.datasetParameters);
~~~~~~~^
2 errors generated.

Can you tell me what might happen and how to fix it? thank you very much

Thanks for your great job!
However, i meet some problems with the source code.
I wonder if this is available in mac osx? There exists some problems compiling master branch:

1. cmake set openmp lib as libomp, however, app clang does not like this
2. MatchCubes.h needs include <Eigen/StdVector>
3. MixedSolverSchurMp,h seems to not find recursive_conjugate_gradient_OMP as _OPENMP not defined?
4. some library need to link to cholmod.

After solving these problems, it finally compiled successfully on my macbook.
However, when running the app, it stucked in glfwCreateWindow and it's hard for me to solve it.

Wish your answer! Thanks.

@darglein
Hi, I would like to ask, as I want to set all tracked frames as keyframe(improve the quality of the reconstruction), and which part of the code should I modify?
thanks