Test data sets for automatic image orientation


Last Update 2007.07.23

Here you can find several data sets for testing automatic image orientation procedures.

These data sets and the ground truth are conceived for testing the robustness of automatic image orientation algorithms. For this purpose the data sets play with repetitive patterns, different light conditions and moving objects.
Run your algorithms on the provided images and compare your exterior orientation parameters with the given "true" orientation parameters. These "true" orientation parameters are derived from manual measurements using the images in the original full resolution. The images provided here have reduced resolution.
The original images are affected by small amounts of non-linear distortion (less than 5 pixel in 90% of the image). Consequently depending on the non-linear distortion parameters used in your algorithm the resulting orientation parameters may slightly deviate from the given "true" ones - even if your results are "correct".

Consequently depending on the non-linear distortion parameters used in your algorithm the resulting orientation parameters may slightly deviate from the given "true" ones - even if your results are "correct". Therefore for each data-set a file checkpoi_halfres.txt with check points is provided. There you will find a set of 3D points and their corresponding image points in all images. These points are virtual and their image mappings are computed using the "true" orientation parameters (and considering the non-linear distortion). By applying your resulting orientation on these image points you get corresponding 3D coordinates which you can compare with the given true ones (after a spatial similarity transformation). So, there are two ways of comparing your results with the given ground truth: (a) comparison of the exterior orientation parameters, (b) comparison of the 3D coordinates of the virtual check points.

This is a first attempt in providing such data sets (generated in the course of ISPRS working group III/1 2004-2008). Therefore any feedback on these data-sets is welcome: car@ipf.tuwien.ac.at

	Data set  Stones 5 images natural object with similar or repetitive patterns real day light conditions	Download zip-archive (2.2MB)
	Data set  Dish ware two sets with 16 images each natural object with similar or repetitive patterns difficult light and surface properties light conditions: one set of images taken under fluorescent light, the other set uses flash light	Download zip-archive (11.1MB)
	Data set  Bric-a-brac six sets with 3-5 images each data aims at testing robustness (moving objects, fixed for groups of images)	Download zip-archive (5.1MB)
	Data set  Covers eight sets with 4 images each data aims at testing robustness (moving objects, different in each image)	Download zip-archive (11.2MB)