In the original testing/height01.py file, the height estimation error is negligible, which is around 0.001 ~ 0.002 meter for all pixels.
However in other test files such as the testing/humanSG02.py, the error goes to 0.1 - 0.2 meter. Further, the error increases when going to the lower part of the image, which clearly indicates that the estimated surface and the real one is no longer the same.
Two possible reason:
The data collection error. Perhaps the camera position changes in the test frame compared to the heigh_estimator parameter calibration frame
The data quality itself. Maybe the depth frame itself has too much noise. This one is less likely since it cannot explain the error pattern along the vertical axis
In the original testing/height01.py file, the height estimation error is negligible, which is around 0.001 ~ 0.002 meter for all pixels.
However in other test files such as the testing/humanSG02.py, the error goes to 0.1 - 0.2 meter. Further, the error increases when going to the lower part of the image, which clearly indicates that the estimated surface and the real one is no longer the same.
Two possible reason: