Raphaël Millière @raphaelmilliere.com · Feb 17 Of course image diffusion models also fail to capture some aspects of the structure of natural images. For example, they fail to capture correct projective geometry. 11/ arxiv.org/abs/2311.17138
2 likes 1 replies
?
Replies
What about video generation models? As far as I know, we simply lack this kind of research at this point. We need interventional studies that try to decode and manipulate internal representations of intuitive physics properties in these models to establish firmer conclusions. 12/