badslam-test passed 11 test but failed 3 tests in a Jetson AGX Xavier about badslam HOT 3 CLOSED

Thank you for reporting these failures. I believe that they are likely more of an issue with the tests rather than an issue with the tested functionality. There is generally some variance in the test results, for example due to differing CUDA block scheduling. It seems that the thresholds used in the tests may be slightly too small to account for that variance.

In detail:

The test Optimization.PCGIntrinsicsOptimizationWithPhotometricResidual reached a difference of -0.0406952, 0.0275574, -0.0578613, -0.114563 to the ground truth intrinsics, but the thresholds on the absolute difference values are 0.03, 0.03, 0.15, 0.15 (only the first value exceeds the threshold slightly).

The test Optimization.AlternatingIntrinsicsOptimizationWithPhotometricResidual reached a difference of -0.0420685, 0.027298, -0.0567017, -0.11319 to the ground truth intrinsics, but the thresholds on the absolute difference values are 0.03, 0.03, 0.15, 0.15 (again, only the first value exceeds the threshold slightly).

The test Optimization.AlternatingDepthDeformationOptimizationWithGeometricResidual exceeded a threshold of 0.01 more significantly with a value of 0.0427, but there is a comment in the test code that this specific value converges very slowly and probably does not matter much.

from badslam.

Thank you for your explanation. I was worried about these 3 to 4 test failures because it's written all tests should passed in your installation instruction. Since it's not matter much, I can continue work on the badslam.

from badslam.

close issue

from badslam.