Collection of Kinect (RGB+D) datasets with 6D ground truth (by the CVPR team @ Technische Universitat Munchen)

Download dataset filesSize
rawlog_rgbd_dataset_freiburg1_360.tgz 0.4Gb
rawlog_rgbd_dataset_freiburg1_desk2.tgz 0.3Gb
rawlog_rgbd_dataset_freiburg1_desk.tgz 0.3Gb
rawlog_rgbd_dataset_freiburg1_floor.tgz 0.7Gb
rawlog_rgbd_dataset_freiburg1_ir_calibration.tgz 0.6Gb
rawlog_rgbd_dataset_freiburg1_plant.tgz 0.7Gb
rawlog_rgbd_dataset_freiburg1_rgb_calibration.tgz 0.9Gb
rawlog_rgbd_dataset_freiburg1_room.tgz 0.8Gb
rawlog_rgbd_dataset_freiburg1_rpy.tgz 0.4Gb
rawlog_rgbd_dataset_freiburg1_teddy.tgz 0.9Gb
rawlog_rgbd_dataset_freiburg1_xyz.tgz 0.5Gb
rawlog_rgbd_dataset_freiburg2_360_hemisphere.tgz 1.4Gb
rawlog_rgbd_dataset_freiburg2_360_kidnap.tgz 0.7Gb
rawlog_rgbd_dataset_freiburg2_coke.tgz 1.3Gb
rawlog_rgbd_dataset_freiburg2_desk.tgz 1.9Gb
rawlog_rgbd_dataset_freiburg2_desk_with_person.tgz 2.5Gb
rawlog_rgbd_dataset_freiburg2_dishes.tgz 1.5Gb
rawlog_rgbd_dataset_freiburg2_flowerbouquet_brownbackground.tgz 1.7Gb
rawlog_rgbd_dataset_freiburg2_flowerbouquet.tgz 1.2Gb
rawlog_rgbd_dataset_freiburg2_ir_calibration.tgz 0.7Gb
rawlog_rgbd_dataset_freiburg2_large_checkerboard_calibration.tgz 1.4Gb
rawlog_rgbd_dataset_freiburg2_large_no_loop.tgz 1.8Gb
rawlog_rgbd_dataset_freiburg2_large_with_loop.tgz 2.8Gb
rawlog_rgbd_dataset_freiburg2_metallic_sphere2.tgz 1.2Gb
rawlog_rgbd_dataset_freiburg2_metallic_sphere.tgz 1.0Gb
rawlog_rgbd_dataset_freiburg2_pioneer_360.tgz 0.4Gb
rawlog_rgbd_dataset_freiburg2_pioneer_slam2.tgz 1.2Gb
rawlog_rgbd_dataset_freiburg2_pioneer_slam3.tgz 0.9Gb
rawlog_rgbd_dataset_freiburg2_pioneer_slam.tgz 1.3Gb
rawlog_rgbd_dataset_freiburg2_rgb_calibration.tgz 1.1Gb
rawlog_rgbd_dataset_freiburg2_rpy.tgz 2.0Gb
rawlog_rgbd_dataset_freiburg2_teddy.tgz 1.1Gb
rawlog_rgbd_dataset_freiburg2_xyz.tgz 2.2Gb
rawlog_rgbd_dataset_freiburg3_structure_texture_near.tgz 0.6Gb
rawlog_rgbd_dataset_freiburg3_structure_texture_far.tgz 0.6Gb
rawlog_rgbd_dataset_freiburg3_structure_notexture_near.tgz 0.5Gb
rawlog_rgbd_dataset_freiburg3_structure_notexture_far.tgz 0.4Gb
rawlog_rgbd_dataset_freiburg3_nostructure_texture_far.tgz 0.3Gb
rawlog_rgbd_dataset_freiburg3_nostructure_notexture_far.tgz 0.2Gb
rawlog_rgbd_dataset_freiburg3_walking_xyz.tgz 0.5Gb
rawlog_rgbd_dataset_freiburg3_walking_static.tgz 0.5Gb
rawlog_rgbd_dataset_freiburg3_sitting_halfsphere.tgz 0.7Gb
rawlog_rgbd_dataset_freiburg3_sitting_xyz.tgz 0.8Gb
rawlog_rgbd_dataset_freiburg3_sitting_static.tgz 0.5Gb
SensorsRGBD, IMU (not in freiburg3), Ground truth
Recorded at:Freiburg (2011-2012)
Additional info:
This dataset is a derived work from the collection [1] published by the CVPR team in the TUM University. According to the original "Creative Commons Attribution" license, this derived work is also released under identical terms. The dataset has been converted into the Rawlog format [2] and published in the aim of it being useful to MRPT users. It can be visualized and manipulated with MRPT tools, available for download at [3]. This tutorial explains how to parse the datasets to emulate real-time applications from an MRPT program: Ubuntu users can also install:
 sudo apt-get install libmrpt-dev mrpt-apps
and then right-click on *.rawlog files to open them with RawLogViewer (see a demo video on YouTube). Note that RGB, depth and accelerometer data are available for each "observation". 3D point clouds haven't been pre-generated to save space, but they can be easily recovered by using "rawlog-edit":
 rawlog-edit -i [INPUT.rawlog] -o [OUTPUT.rawlog] --generate-3d-pointclouds
Please, cite the original works by the corresponding authors when citing this dataset [4,5]. To download all the datasets in a row, download this list and invoke:
  wget -i rgbd_tum_datasets_2011.txt
The source code of the tool used to convert these datasets into the "rawlog" format is available as one of MRPT's code examples. Some notes about the differences between Kinect datasets in Rawlog format and in the TUM originals:
  • In MRPT, a Kinect observation object (CObservation3DRangeScan) can store the three fields simultaneously: range image, RGB data and projected 3D point coordinates.
  • An observation is created only when both range & RGB data are available in the dataset with a difference in timestamp below half the kinect period (1/30 sec). That means that due to ocasional losses in the original datasets, the rate of observations is less than 30Hz.
  • Only one IMU entry is kept per RGB+Range pair. If you need high-rate accelerometer data, please download the TUM original data files.


[1] Original datasets: [2] Rawlog format specification: [3] MRPT downloads: [4] [5]