we present Iris, a Priors-to-Geometry Deterministic framework that injects real-world priors into the diffusion model for monocular depth estimation. Our two-stage schedule separates prior alignment at a high timestep, where Spectral-Gated Distillation transfers low-frequency real priors, from geometry refinement at a low timestep, where Spectral-Gated Consistency enforces high-frequency agreement under an over-activation constraint. Iris preserves fine details, generalizes strongly from synthetic to real scenes, and remains efficient with limited training data, achieving significant improvements across diverse real-image benchmarks and outperforming both prior diffusion-based approaches and large-scale deterministic feed-forward models.
This installation was tested on: Ubuntu 20.04 LTS, Python 3.10, CUDA 12.3, NVIDIA A100-40GB.
- Install dependencies (requires conda):
conda create -n Iris python=3.10 -y
conda activate Iris
pip install -r requirements.txt
- Download the Iris model from Hugging Face and put it in the
trained_models/
- Place your images in a directory, for example, under
assets/in-the-wild_example(where we have prepared several examples). - Run the inference command:
bash infer.sh. - You can simultaneously view the outputs from both Stage 1 and Stage 2.
- Prepare benchmark datasets:
- Download the evaluation datasets (depth) by the following commands (referred to Marigold):
cd datasets/eval/depth/ wget -r -np -nH --cut-dirs=4 -R "index.html*" -P . https://share.phys.ethz.ch/~pf/bingkedata/marigold/evaluation_dataset/
- Run the evaluation command:
bash eval_scripts/eval.sh
If you find this repository useful, please consider citing:
@article{cai2026iris,
title={Iris: Bringing Real-World Priors into Diffusion Model for Monocular Depth Estimation},
author={Cai, Xinhao and Pei, Gensheng and Sun, Zeren and Yao, Yazhou and Shen, Fumin and Wang, Wenguan},
journal={arXiv preprint arXiv:2603.16340},
year={2026}
}