rather than just the intrinsics. This still means that the data was wrong in the first place and affects all other algorithms, as most used just the intrinsics.
In addition, it's been determined that the DepthSensorSerialiser and DepthSensorDeserialiser classes ignored distortion parameters and did not write or read this from SLAMFile, unlike CameraSensorSerialiser. Furthermore, in TUM.cpp in GenerateSLAMFile, loops do not capture the last distortion parameter.
Noticed while testing ORB-SLAM2 on the TUM dataset.
Camera parameters in
framework/tools/dataset-tools/include/TUM.h
do not match the description in https://vision.in.tum.de/data/datasets/rgbd-dataset/file_formats, and the values are unrealistic.The reason ORB-SLAM2 still works is that the values stored in the file actually result in the correct values when using
rather than just the intrinsics. This still means that the data was wrong in the first place and affects all other algorithms, as most used just the intrinsics.
In addition, it's been determined that the
DepthSensorSerialiser
andDepthSensorDeserialiser
classes ignored distortion parameters and did not write or read this from SLAMFile, unlikeCameraSensorSerialiser
. Furthermore, inTUM.cpp
inGenerateSLAMFile
, loops do not capture the last distortion parameter.