-
Notifications
You must be signed in to change notification settings - Fork 84
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
ptam assertion failure #8
Comments
mihaibujanca
added a commit
to mihaibujanca/slambench2
that referenced
this issue
Oct 13, 2020
* Comment out CUDA monitoring * Add BundleFusion branch code * Fix Makefile missing \ in OpenLORIS * README: remove quick start for IROS challenge The information of lifelong_loader and the new relocalize API is added into proper places. Also corrected the definition of other APIs. Signed-off-by: Xuesong Shi <[email protected]> * add PCL to framework/shared/CMakeLists to fix compiling issue Signed-off-by: Xuesong Shi <[email protected]> * remove unused code introduced by this PR Signed-off-by: Xuesong Shi <[email protected]> * fix depth scaling factors of TUM/BONN/ICL-NUIM The [TUM file formats] (https://vision.in.tum.de/data/datasets/rgbd-dataset/file_formats) states that depth images are scaled by a factor of 5000. However, the factor had been set to 1000 in slambench2 (disparity_params=0.001) because there was a hardcoded dividing by 5 in SLAMFrame.cpp. Now that we had removed the dividing by 5, we are fixing the scaling factors for TUM and other TUM-format datasets. Signed-off-by: Xuesong Shi <[email protected]> * fix depth scaling factor of ICL Following last commit Signed-off-by: Xuesong Shi <[email protected]> * lifelong_loader: align traj only with gt of the first input For lifelong SLAM evaluation, the estimated poses should be aligned with groundtruth of the first data input. This commit implements this feature by introducing AlignmentOutput::SetFreeze(), which will be called by SLAMBenchConfigurationLifelong at the end of the first input. Then, AlignmentOutput will stop updating the transformation matrix. Also added the alignment-technique (-a) param to lifelong_loader. And fixed param explanation in benchmark_loader. Signed-off-by: Xuesong Shi <[email protected]> * support dataset-generator options for OpenLORIS in Makefile Signed-off-by: Xuesong Shi <[email protected]> * remove SLAMBenchLibraryHelperLifelong This was introduced to keep minimal change to SLAMBenchLibraryHelper. But as libraries should be re-compiled anyway, it is unnecessary to maintain a 'lifelong' lib helper other than the 'normal' one, which would be another historical strangeness. BTW, the getSensorUpdate() API is no longer needed by the libraries. (since the workaround of reusing input interface in SLAMBenchConfigurationLifelong::AddInputInterface) Signed-off-by: Xuesong Shi <[email protected]> * some minor code improvements Signed-off-by: Xuesong Shi <[email protected]> * Add OpenLORIS targets to .PRECIOUS * Set the default alignment method to "new" on all loaders * Remove unused headers in OpenLORIS.cpp * fix association and implement interpolation The previous implementation of associate() made an assumption that gt traj is 'denser' than the estimated traj. When it is violated, e.g. with the OpenLORIS home1 data, runtime error will occur. This commit refactored the associate() function to remove the assumption. It also implements interpolation between two closest gt poses. Signed-off-by: Xuesong Shi <[email protected]> * fix groundtruth mismatching in SLAMBenchConfiguration This is a quick workaround and I'm expecting more elegent fixings. The implementation of GTBufferingFrameStream looks over-complicated and confusing to me. I think it made an assumption that gt should come first from the deserialiser, and the buffer works only once at initilization (perhaps should call fastForward in GetNextFrame?). And it may be possible to make a more straightforward implementation of GetGTFrames. Signed-off-by: Xuesong Shi <[email protected]> * fix crash when groundtruth ends earlier than other frames Keep calm, print nan, and carry on. Signed-off-by: Xuesong Shi <[email protected]> * change raw ptr to smart ptr for Metrics and refine cw management Signed-off-by: Xuesong Shi <[email protected]> * refactor result saving and move it out from SLAMBenchConfigurationLifelong So that other loaders could re-use it. Also added saving groundtruth, to faciliate post analysis. Signed-off-by: Xuesong Shi <[email protected]> * OpenLORIS builder: transform gt target to d400_color And changed sensor poses accordingly. Though nobody cares. Require to re-build all .slam file in OpenLORIS. Signed-off-by: Xuesong Shi <[email protected]> * add absolute orientation error into ATEMetric Signed-off-by: Xuesong Shi <[email protected]> * SLAMBenchConfiguration(Lifelong): consistent member naming This does not change any program behavior. Signed-off-by: Xuesong Shi <[email protected]> * OpenLORIS.h/.cpp: format code This does not change any program behaviors. Signed-off-by: Xuesong Shi <[email protected]> * copy evaluate scripts from openloris_tools Signed-off-by: Xuesong Shi <[email protected]> * Minor code improvements Use auto when possible, change pass by value to const reference * process gt pose properly in post analysis Signed-off-by: Xuesong Shi <[email protected]> * Fix unused parameter error * Remove open_vins directory Co-authored-by: Mihai Bujanca <[email protected]> Co-authored-by: Xuesong Shi <[email protected]>
seeing the same issue (ubuntu 20.04) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
algorithm: libptam-original_mp-library
platform: x86_64, ubuntu 16.04
build: Thursday, April 12, 12:50
Assertion failure and return non-zero
The text was updated successfully, but these errors were encountered: