M^3: Dense Matching Meets Multi-View Foundation Models for Monocular Gaussian Splatting SLAM
Paper • 2603.16844 • Published • 9
M³ matching checkpoints
Matching weights for M³: Dense Matching Meets Multi-View Foundation Models for Monocular Gaussian Splatting SLAM.
Accepted to SIGGRAPH Asia 2026.
- Code: https://github.com/InternRobotics/M3
- Paper: https://arxiv.org/abs/2603.16844
Pi3.pt: original Pi3 matching checkpoint (pi32).Pi3X.pt: original Pi3X matching checkpoint (pi3x3).
These are original checkpoint files, not converted or quantized weights. They are
matching models, not interchangeable with the unmodified base geometry models.
Use the corresponding Pi3/Pi3X matching configuration. The state dictionary is
stored under the model key; use strict loading and load only trusted files.
See SHA256SUMS and weights.json for integrity and exact weight revisions.
The cleaned code release and fresh CUDA regression are still being prepared.
No blanket license for third-party components is granted by this model card;
upstream model terms remain applicable.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support