I understand that each model is normalized to fit within a unit cube centered at the origin. I verified this by checking several .obj files and find the vertice coordinates are all between 0.5 and -0.5. However, the min/max value of the coordinates does not equal -0.5/0.5, and the origin does not match the center of mass. Is there a normalization rule shared among the dataset? or among objects in the same category?
I think I found the answer. I misunderstood the explanation in README.
The norm of (max(x)-min(x), max(y)-min(y), max(z)-min(z)), which I think is the diagonal distance of the bounding box, is close to 1. I also found that (max(x)+min(x))/2, (max(y)+min(y))/2, (max(z)+min(z))/2 are all close to 0, which means the center of bounding box is aligned to the origin. This should be the normalization rule.