I ran 10 batch tests giving alignment scores for the 1161 blobs against the 42 meshes. I then created an ARFF file for each test. The order of the blobs tested is presented from top to bottom in this file. They were tested against all of the meshes in the order presented in this file.
The name of the arff file indicates what type of batch test was run. In each test, I sampled a maximum of 100 points from the point cloud and did ICP alignment to each of the 42 mesh models with that subset. Then I chose an cutoff epsilon to say whether the aligned points were close enough to the mesh (the ratio of close enough points is the score), and I chose to do either 6D or 3D. For example, take the arff file test100_0.200.arff. This file means that 6D alignment was done with a cutoff epsilon of 0.2 meters. The file test100_3D_0.050.arff means that 3D alignment was done with a cutoff epsilon of 0.05 meters.
Once I generated an ARFF file, I did Naive Bayes using Weka and got the probability matrix for classification. I stored this probability matrix in a text file, and I created precision recall graphs from it. This process is described in more detail here.
Click here to download all of the arff files and the Naive Bayes probability matrices for each arff file.