Last-Layer-Centric Feature Recombination: Unleashing 3D Geometric Knowledge in DINOv3 for Monocular Depth Estimation

Benchmark Model Rank Results
depth-estimation-on-nyu-depth-v2DINOv3-B+LFR–RMS: 0.251
depth-estimation-on-nyu-depth-v2DINOv3-L+LFR–RMS: 0.206
monocular-depth-estimation-on-kitti-eigen-splitDINOv3-L+LFR–Delta < 1.25: 0.988
monocular-depth-estimation-on-nyu-depth-v2DINOv3-B+LFR–absolute relative error: 0.070RMSE: 0.251log 10: 0.030…
monocular-depth-estimation-on-nyu-depth-v2DINOv3-L+LFR–absolute relative error: 0.057RMSE: 0.206log 10: 0.025…