paulmrinal
commented
6 years ago Hi,
Sorry for your trouble. The drivers on this branch were experimental and might have some bugs. I will have a look into it. However, I am currently a bit pre-occupied and it might take some time before I am able to fix the bug.
In the mean time, if you are ok with having the absolute RMSE data, I am listing them here.
| Dataset | Absolute RMSE (m) |
|---|---|
| MH_01_easy | 0.17501 |
| MH_02_easy | 0.063584 |
| MH_03_medium | 0.098886 |
| MH_04_difficult | 0.3326 |
| MH_05_difficult | 0.309 |
| V1_01_easy | 0.066619 |
| V1_02_medium | 0.055443 |
| V1_03_difficult | 0.09485 |
| V2_01_easy | 0.047206 |
| V2_02_medium | 0.071117 |
The results are using this configuration: max_number_of_extracted_keypoints = 400, sliding_window_size = 5, image_processing_rate = 30 Hz (to match the default configuration of OKVIS). The released binaries, however, have different configuration (max_number_of_extracted_keypoints = 600, sliding_window_size = 10, image_processing_rate = adaptive, parameters tuned for wide FOV fisheye cameras), so the results will vary.
P.S. The V2_03_difficult dataset seemed to have some issues (do not remember what exactly was the problem), so I did not include it in my results.
wtrobin
Hi,
I've tried to evaluate your tracker's performance on the (stereo) EuRoC data set. So far I've converted all the images (.png -> .pgm), converted the IMU data file to comply to the MARS-VINS format, and I've adapted the sample_config.txt to the EuRoC camera calibration and IMU values.
After a few tiny tweaks in the generic_driver.cpp (to be able to read EuRoC images instead of m0000012.pgm) I'm finally able to start the example application. However, it crashes somewhere inside the binary and I'm not sure what I can do about that.
The attached Valgrind output highlights two mismatched new/delete pairs (during image loading). The crash is caused by an invalid read in Estimators::StereoInitializerEngine::set_imu_measurement()
Any help would be appreciated.
Thanks, Werner
P.S. If fixing this is tricky, I'd also be happy with absolute trajectory errors (RMSE) for all individual EuRoC sequences :-)