NVIDIA announced MetricTrail, a tool for visual odometry and 3D reconstruction research. It estimates camera motion and per-frame metric depth from long RGB image sequences. MetricTrail 1.0 was developed by NVIDIA and is available on Hugging Face.
Key facts
| Fact | Detail | The source says |
|---|---|---|
| Use Case | Research on long-video camera-motion estimation and metric 3D reconstruction | “Research on long-video camera-motion estimation and metric 3D reconstruction.” |
| Release Date | 11/05/2026 | “Hugging Face 11/05/2026 via https://huggingface.co/nvidia/metrictrail” |
| Model Architecture | Learned recurrent visual odometry with frozen visual foundation-model features and explicit bundle adjustment | “Architecture Type: Learned recurrent visual odometry with frozen visual foundation-model features and explicit bundle adjustment.” |
What happened
NVIDIA released MetricTrail, a tool for visual odometry and 3D reconstruction research. It estimates camera motion and per-frame metric depth from long RGB image sequences. MetricTrail combines single-image metric depth, learned sparse correspondences, and local geometric optimization, with supplied or RGB-estimated intrinsics and optional rectified stereo. The repository provides a command-line interface (CLI) that exports camera trajectories and caches per-frame depth. Model weights are not included in the source repository; the DINO-DPVO checkpoint is planned for a separate Hugging Face release. The CLI does not export dense point clouds or meshes. The dense-reconstruction results use a separate research evaluation workflow. This model is for research and development only.
What to weigh
- MetricTrail is intended for research and development purposes only.
- Model weights are not included in the source repository.
FAQ
What is the use case for MetricTrail?
MetricTrail is for research on long-video camera-motion estimation and metric 3D reconstruction.
Is MetricTrail available for commercial use?
MetricTrail is intended for research and development only.
Where can I find the model weights?
The DINO-DPVO checkpoint will be released separately on Hugging Face.
Source: NVIDIA
