Hi,
After over 3000 runs of the experiment I think I have reached the source of the problem.
Both stimuli (large circles with small dots) have a mask that hides their background, and of course is the same mask and the same file, when I remove from one of them the mask (which is not actually necessary for the experiment) has no crash.
Maybe the source of the problem was the overlap between the masks? Maybe an overlap between two masks that are the same file? I do not know. But a solution, at the moment, I have found.