DepthCrafter generates temporally consistent, high-fidelity depth sequences for open-world videos without requiring camera poses or optical flow. It is targeted at researchers and developers in computer vision and VFX, offering improved quality and speed over existing methods.
How It Works
DepthCrafter employs a novel approach to produce long, consistent depth sequences by leveraging a diffusion model trained on extensive video data. This method inherently handles temporal coherence and fine-grained details, eliminating the need for explicit motion estimation or camera pose information, which simplifies the pipeline and broadens applicability.
Quick Start & Requirements
- Install:
git clone https://github.com/Tencent/DepthCrafter.git
followed by pip install -r requirements.txt
.
- Prerequisites: GPU with ~26GB memory for high-resolution (1024x576) inference, or ~9GB for lower resolutions (512x256).
- Demo: Gradio demo available online, or run locally with
gradio app.py
.
- Resources: Project page for visualizations.
Highlighted Details
- Selected as a Highlight at CVPR '25.
- Achieves state-of-the-art performance on multiple benchmarks (Sintel, ScanNet, KITTI, Bonn), outperforming Depth-Anything-V2 and Marigold in AbsRel and δ₁ metrics.
- Supports EXR output format.
- Integrated into Nuke and ComfyUI for professional VFX and creative workflows.
- Offers a Hugging Face online demo.
Maintenance & Community
- Actively under development with v1.0.1 released for improved quality and speed.
- Community support via GitHub issues. Related nodes for Nuke and ComfyUI are available.
- Business inquiries can be directed to wbhu@tencent.com.
Licensing & Compatibility
- The repository does not explicitly state a license. Users should verify licensing for commercial use or closed-source integration.
Limitations & Caveats
- The project is still under active development, indicating potential for breaking changes.
- Communication is recommended in English for broader community support.
- No explicit license is provided, which may pose a barrier for commercial adoption.