Marigold V2: Diffusion Transformer Becomes SOTA Monocular Depth Estimator With Single-Step Inference

AntonObukhov1 · x · 2026-09-12

Marigold V2 is out: a diffusion transformer fine-tuned into a state-of-the-art monocular depth estimator.

Key selling point is efficiency — it supports single-step inference, and training is possible on a single consumer GPU, sharply lowering the barrier for researchers. Released via the Hugging Face ecosystem.

Related event: Marigold V2 Achieves SOTA Monocular Depth Estimation in a Single Step(2 posts)→

Original post →

More from Research

Research channel →