I imagine this is based on their 'Thinking with Visual Primitives' paper, and they had mentioned that the weights would be released for that