The TUT head pose competition was organized on Fall 2014. The task was to recognize the head pose from still images of the Pointing'04 database. Dense HOG features are provided by the organizers and calculated using the vlfeat toolbox.

The results of the competition are reported in our paper:
The competition data can be downloaded from this link. The archive contains the following five files:
  • X_train.csv: a csv file containing all extracted features for the 1953 training samples.
  • y_train.csv: a csv file containing yaw and pitch angles for the 1953 training samples.
  • X_test.csv: a csv file containg all extracted features for the test samples.
  • y_scoring.csv: a csv file containg yaw and pitch angles for the test samples. The last column defines the use of each sample: "Public" means that the competitors were allowed to probe the accuracy of their method against this data daily. "Private" samples are used only at the end of the competition, and the final results (and all experiments reported in our paper) are based on private samples. You can consider the "Public" samples as a validation set and "Private" as an independent test set.
  • trainRidge.m: An example script that predicts angles using ridge regression.

If you use the data, please cite the above paper.