2K Retrofit enables 2K-resolution inference for any 3D geometric foundation model without modifying or retraining the backbone.
March 23, 2026
Original Paper
2K Retrofit: Entropy-Guided Efficient Sparse Refinement for High-Resolution 3D Geometry Prediction
arXiv · 2603.19964
The Takeaway
Modern geometric models (depth, normals) are capped by resolution due to memory. This framework uses entropy-guided sparse refinement to selectively enhance high-uncertainty regions, allowing practitioners to achieve high-fidelity 2K outputs at the speed of low-res models.
From the abstract
High-resolution geometric prediction is essential for robust perception in autonomous driving, robotics, and AR/MR, but current foundation models are fundamentally limited by their scalability to real-world, high-resolution scenarios. Direct inference on 2K images with these models incurs prohibitive computational and memory demands, making practical deployment challenging. To tackle the issue, we present 2K Retrofit, a novel framework that enables efficient 2K-resolution inference for any geome