How do you split the training and testing sets?

[zhimi9999]

Hello, I am developing a grasping algorithm using your dataset. I would like to ask how are the training and test sets in your data set divided?

[mzhmxzh]

train.txt 4229 objects test.txt 1126 objects This is how we divided the dataset in our benchmark experiments. But you can also divide them differently.

[zhimi9999]

Did you merge all the datasets and randomly shuffle them? I have a strange result that the performance on the test set is actually better than that on the training set.

[mzhmxzh]

The train/test sets are indeed divided randomly, except for some categories that contain only a few instances, which are all sorted to the test set. So I cannot explain why your algorithm has a higher performance on the test set...