Strange translation values for urban05 dataset
BorSch90 opened this issue · comments
Dear @simone-fontana,
thank you for the dataset which I have found to be very usable and diverse!
The urban05 (global and local) transforms seem to have strange translation values, e.g.
pose1 = (-2246611.1391324443, 9356676.980349164,-7403041.313363417), pose2 = ( -6592076.38153361, 8562547.723609477, -3427421.61174382). Is there a way to fix them and make them metric?
See https://github.com/iralabdisco/point_clouds_registration_benchmark/blob/1d42c6db2571cae960c45dad52ba12c09642c04c/devel/registration_pairs/urban05_global.txt#L2C1-L2C1
Hi!
The ground truth of the original urban05 dataset was provided in UTM coordinates, this is the reason behind the very large numbers of the translations. They actually are metric coordinates (as per UTM specs), just very large.
Based on our experience they should be not too large to cause loss of precision.
One way to have smaller translation would be to "shift" the origin of reference system of the whole sequence. That is, subtract the same (large) 3d vector from every translations and from the point clouds.
UTM -> I see, makes sense! I also do not expect loss of precision from doubles.
I am not quite sure how substracting any vector from e.g. pose1 = (-2246611.1391324443, 9356676.980349164,-7403041.313363417), pose2 = ( -6592076.38153361, 8562547.723609477, -3427421.61174382) would lead to an improvement though? Those things are millions of meters apart?