Distill-Any-Depth: Monocular Depth Estimator

1. Tutorial Introduction

Distill-Any-Depth is an innovative monocular depth estimation project jointly released by Zhejiang University of Technology, Westlake University, Henan University, and the National University of Singapore on February 28, 2025. The project integrates the advantages of multiple open source models through the distillation algorithm, and can achieve high-precision depth estimation with only a small amount of unlabeled data, refreshing the current SOTA (State-of-the-Art) performance.Distill Any Depth: Distillation Creates a Stronger Monocular Depth Estimator".

Here are its key points:

• Multi-Teacher Distillation Framework
  • By randomly selecting multiple teacher models to generate pseudo labels, the advantages of different models are combined to improve the quality of pseudo labels.
  • The cross-context distillation mechanism is introduced to combine local details with global information, significantly enhancing the robustness of the model.
• Local Normalization Strategy
  • Traditional global normalization will amplify noise. This project proposes to perform normalization within the cropped area to retain local details (such as object edges and small hole structures) and improve prediction accuracy.
• Low data dependency
  • Only 20,000 unlabeled images are needed (far lower than the millions of annotations required by traditional methods), which greatly reduces the cost of data annotation.
• Generalization
  • In benchmark tests such as NYUv2 (indoor), KITTI (outdoor driving), and DIODE (complex lighting), the error indicator (AbsRel) is significantly better than the previous model.
• robustness
  • It performs stably in transparent objects, reflective surfaces and dynamic scenes, solving the failure problem of traditional models under complex conditions.
• efficiency
  • The inference speed is more than 10 times faster than that of Diffusion-based models (such as Marigold), supporting real-time applications.

The computing resources used in this tutorial are a single RTX 4090 card.

2. Effect display

3. Operation steps

1. Start the container

If "Bad Gateway" is displayed, it means the model is initializing. Since the model is large, please wait about 2-3 minutes and refresh the page.

2. Usage steps

result 

4. Discussion

🖌️ If you see a high-quality project, please leave a message in the background to recommend it! In addition, we have also established a tutorial exchange group. Welcome friends to scan the QR code and remark [SD Tutorial] to join the group to discuss various technical issues and share application effects↓

Citation Information

The citation information for this project is as follows:

@article{he2025distill,
  title   = {Distill Any Depth: Distillation Creates a Stronger Monocular Depth Estimator},
  author  = {Xiankang He and Dongyan Guo and Hongji Li and Ruibo Li and Ying Cui and Chi Zhang},
  year    = {2025},
  journal = {arXiv preprint arXiv: 2502.19204}
}