Search papers, labs, and topics across Lattice.
FoundationGeo introduces a two-stage framework that enhances monocular metric geometry by integrating relative and metric predictions through spatial calibration and data design. The first stage employs a high-fidelity geometry model trained on a large, curated dataset, achieving strong cross-domain generalization and sharp boundaries. The second stage incorporates pixel-wise calibration fields to address directional biases and improve metric consistency, resulting in superior zero-shot performance across multiple benchmarks, outperforming heavier models by over 5.2%.
By synthesizing diverse training data and leveraging pixel-wise calibration, FoundationGeo achieves unprecedented robustness in monocular metric geometry, even under varying camera intrinsics.
We present FoundationGeo, a two-stage framework that explicitly bridges relative and metric prediction via spatial calibration and principled data design. Stage 1 learns a high-fidelity, affine-invariant geometry model by initializing with DINOv3 and training on a curated 10.2M-sample multi-domain corpus with complementary local-detail supervision, yielding sharp boundaries and strong cross-domain generalization. Stage 2 moves beyond global scaling by introducing lightweight pixel-wise calibration fields for metric estimation: a scale field for spatially varying metric alignment and a ray-direction correction field that mitigates directional bias in point-map geometry, together producing metrically consistent 3D point maps. Beyond model design, we identify camera intrinsic coverage, especially focal length distribution mismatch between training and test data, as a key bottleneck for zero-shot metric generalization: performance drops sharply when test intrinsics fall outside the training distribution. To address this, we synthesize additional training data across diverse focal lengths using a Blender-based data engine, repairing under-covered focal regimes and improving robustness under intrinsic shift. Extensive zero-shot evaluations across seven benchmarks show that FoundationGeo significantly strengthens cross-domain robustness, staying near the top across diverse domains while avoiding the sharp cross-domain performance drops observed in other methods. This consistency translates into the best overall performance, surpassing heavier baselines by over 5.2% on average.
