Efficient Image Synthesis with Sphere Latent Encoder
arXiv preprint, 2026
Few-step image generation by decoupling encoding from denoising in a spherical latent space.
Tung Do
I am currently a master’s student in Computer Vision at MBZUAI, under the supervision of Prof. Hao Li. I was a research resident at Qualcomm AI Research (formerly VinAI), advised by Prof. Binh-Son Hua (primary) and Dr. Rang Nguyen (secondary).
Before that, I received a B.Eng. in Computer Engineering from the University of Information Technology - VNUHCM.
News
-
Jul 2026Sphere Latent Encoder is out! We show that few-step image generation becomes more efficient and stable by decoupling a fixed image encoder from a latent denoiser trained entirely on the hypersphere — outperforming Sphere Encoder on ImageNet-1K, Animal-Faces, and Oxford-Flowers.
-
May 2025I will be at CVPR 2025 (Nashville) to present SharpDepth.
-
May 2025I will join MBZUAI to pursue a Master’s degree in Computer Vision.
-
Apr 2025Started as a research resident at Qualcomm AI Research.
-
Feb 2025SharpDepth has been accepted at CVPR 2025.
-
Oct 2024SharpDepth has been released.
-
Jun 2024LiftRefine has been released.
-
Feb 2023Joined VinAI Research as a research resident.
Publications
* equal contribution
SharpDepth: Sharpening Metric Depth Predictions Using Diffusion Distillation
CVPR 2025
Harmonizing diffusion-based depth estimation and metric depth estimation.
LiftRefine: Progressively Refined View Synthesis from 3D Lifting with Volume-Triplane Representations
AI for Content Creation Workshop @ CVPR 2025
Large reconstruction model with 2D diffusion model.