NVIDIA Publishes MetricTrail 1.0 Research Model on Hugging Face

Model Release

NVIDIA's MetricTrail 1.0 is available on Hugging Face under the nvidia-open-model-license. The model card lists the Hugging Face release date as 11/05/2026. The model is for research and development only, and use of the model weights is governed by the NVIDIA License, which limits use to noncommercial research or evaluation, subject to the agreement.

What the Model Does

MetricTrail estimates camera motion and per-frame metric depth from long RGB image sequences for visual odometry and 3D reconstruction research. It combines single-image metric depth, learned sparse correspondences, and local geometric optimization, with supplied or RGB-estimated intrinsics and optional rectified stereo.

The architecture uses DPVO's sparse patch correspondence network with learned residual feature-fusion adapters, a frozen DINOv3-S/16 encoder, and a frozen MoGe-2 metric-depth provider. The DINO-DPVO checkpoint contains 3,425,860 learned parameters (3.43 million) across 100 FP32 weight/bias tensors, excluding the frozen DINOv3 backbone, external MoGe-2 depth model, calibration dependencies, and optional SALAD and RoMa models.

Training and Outputs

MetricTrail was trained for 15,000 optimizer steps with a global batch size of 16 on 16 NVIDIA A100 80GB GPUs, taking approximately seven hours. Training used 240,000 sampled clip exposures across five synthetic sources: TartanAir, TartanGround, CarlaOcc, LightwheelOcc, and OmniWorld-Game.

The CLI exports camera trajectories and caches per-frame depth. It does not export dense point clouds or meshes. Outputs include per-frame camera trajectories, intrinsics, cached frame-level metric depth and provider masks, and processing and audit metadata. Trajectories use meters for translation, and cached depth uses meters along the camera Z axis.

Availability and Requirements

Model weights are not included in the source repository. The DINO-DPVO checkpoint is planned for a separate Hugging Face release. The pipeline requires CUDA, with the documented environment being Python 3.10, PyTorch 2.5.1, and CUDA 12.4. Linux x86_64 is the documented tracking environment, and DPVO CUDA extensions must be built from source before tracking.

Sources

• huggingface.co: nvidia/MetricTrail · Hugging Face

This article was drafted with AI assistance and checked against the sources above. Company claims are reported as claims. Cover image is AI-generated.