Haofei Xu(@haofeixu):Does 3D reconstruction have to be complex? We answer this question with PointDiT (#ICML2026): a minimalist pixel-space Diffusion Transformer without bells and whistles. We show that a plain ViT can estimate dense 3D point maps by operating directly on raw patches. No hybrid ViT+Conv architectures, no lossy VAEs, no complicated training losses. (1/5) 🧵👇

Haofei Xu

@haofeixu

PhD student @ETH_en and @uni_tue. Training neural networks to solve 3D, 4D, and world models.

加入 February 2022

758 正在關注 1.3K 粉絲

Haofei Xu@haofeixu

2026.07.03 10:43

Does 3D reconstruction have to be complex? We answer this question with PointDiT (#ICML2026#): a minimalist pixel-space Diffusion Transformer without bells and whistles. We show that a plain ViT can estimate dense 3D point maps by operating directly on raw patches. No hybrid ViT+Conv architectures, no lossy VAEs, no complicated training losses. (1/5) 🧵👇