The following mat file contains the attribute classifier scores, which allow reproducing Table 5 "All" Column in Table 5 "Fine-grained activity and object classification results" of our [IJCV 15] paper.

1. a list which shows the IDs and names of the different experiments. this corresponds to the "All" Column in Table 5 "Fine-grained activity and object classification results", they are also listed below.
http://datasets.d2.mpi-inf.mpg.de/MPII-Cooking-2/intermediteResults/TrainVal-Test/attributes.txt

2. The second file is
http://datasets.d2.mpi-inf.mpg.de/MPII-Cooking-2/intermediteResults/TrainVal-Test/attributes_scores.mat (101 MB)

It contains four variables, each has 10 cells, which correspond to the indexes above, the one you are probably mainly interested is the last, i.e. index 10.

The scores provide the scores for the attributes (and composite activities for the larger IDs)
‘scores_training'
'scores_test'

to match them to the semantic of the annotations (see http://www.mpi-inf.mpg.de/departments/computer-vision-and-multimodal-computing/research/human-activity-recognition/mpii-cooking-2-dataset/)
you need the following to variables, which are binary indexes into the the annotations. e.g. if training_index is 0 it means that this index of annotations was not used for training and thus is not in the scores_training.
'training_index’, 
‘test_index'

=================

1)      8023.0                  15.7    BM
2)      8023.0                  17.2    FFT
3)      8023.0                  20.8    BM+FFT
4)      8023.0                  23.5    Hand-cSift
5)      8023.0                  36.4    Hand-Trajectories
6)      8023.0                  37.5    Hand-Trajectories + Hand-cSift
7)      8023.0                  36.1    Dense Trajectories
8)      8023.0                  35.2    Dense Traj + BM + FFT
9)      8023.0                  42.9    Dense Traj + Hand-Traj
10)     8023.0                  43.7    Dense Traj + Hand-Traj + Hand-cSift

=================
[IJCV 15] Recognizing Fine-Grained and Composite Activities using Hand-Centric Features and Script Data, M. Rohrbach, A. Rohrbach, M. Regneri, S. Amin, M. Andriluka, M. Pinkal, B. Schiele, IJCV 2015
@article{rohrbach15ijcv, 
year={2015}, 
journal={International Journal of Computer Vision}, 
title={Recognizing Fine-Grained and Composite Activities Using Hand-Centric Features and Script Data}, 
author={Rohrbach, Marcus and Rohrbach, Anna and Regneri, Michaela and Amin, Sikandar and Andriluka, Mykhaylo and Pinkal, Manfred and Schiele, Bernt}, 
issn={0920-5691}, doi={10.1007/s11263-015-0851-8}, url={http://dx.doi.org/10.1007/s11263-015-0851-8}, publisher={Springer US}, pages={1-28}, }