Pixel-Perfect Depth with Semantics-Prompted Diffusion Transformers

Guang Chen (Tongji University) · Xin Yang (Huazhong University of Science and Technology) · Gangwei Xu (Huazhong University of Science and Technology) · Haotong Lin (Zhejiang University) · Hongcheng Luo (Xiaomi Corporation) · Xianqi Wang (Huazhong University of Science and Technology) · JINGFENG YAO (Huazhong University of Science and Technology) · Lianghui Zhu (Huazhong University of Science and Technology) · Yuechuan Pu (Horizon Robotics) · Cheng Chi_ (Huazhong University of Science and Technology, Tsinghua University) · Haiyang Sun · Bing Wang (Alibaba Group) · Hangjun Ye (Xiaomi Corporation) · Sida Peng (Zhejiang University)
cascade dit designdiffusion generationedge-aware point cloud evaluationflying-pixel-free point cloudsgenerative depth estimationlatent spacemonocular depth estimationpixel-perfect depthpixel-space diffusion generationsemantic representationssemantics-prompted diffusion transformerssp-ditstable diffusionvae

This paper presents **Pixel-Perfect Depth**, a monocular depth estimation model based on pixel-space diffusion generation that produces high-quality, flying-pixel-free point clouds from estimated depth maps. Current generative depth estimation models fine-tune Stable Diffusion and achieve impressive performance. However, they require a VAE to compress depth maps into the latent space, which inevitably introduces flying pixels at edges and details. Our model addresses this challenge by directly performing diffusion generation in the pixel space, avoiding VAE-induced artifacts. To overcome the high complexity associated with pixel-space generation, we introduce two novel designs: 1) **Semantics-Prompted Diffusion Transformers** (**SP-DiT**), which incorporate semantic representations from vision foundation models into DiT to prompt the diffusion process, thereby preserving global semantic consistency while enhancing fine-grained visual details; and 2) **Cascade DiT Design** that progressively increases the number of tokens to further enhance efficiency and accuracy. Our model achieves the best performance among all published generative models across five benchmarks, and significantly outperforms all other models in edge-aware point cloud evaluation. Project page: https://pixel-perfect-depth.github.io/.