Same as loftr one, 2px opencv MAGSAC etc.

DPT-KITTI & RoMa (Ess.Mat. + D.Scale) Note: maybe buggy

We replace the baseline DPT-KITTI with KBR depth predictions. KBR is a self-supervised monocular depth estimation system trained on a combination of SlowTV, MannequinChallenge & Kitti. The model is transferred to MapFreeReloc in a zero-shot fashion, without fine-tuning. 

Since KBR is trained only on monocular data (ie without a stereo baseline), the predictions are scaleless. Metric-scale predictions are obtained by aligning KBR to the provided baseline DPT-KITTI using least-squares.

KBR & LoFTR (PnP)

DPT-KITTI & LoFTR (PnP)

DPT-KITTI & LoFTR (Ess.Mat. + D.Scale)

DPT-KITTI & SuperGlue (Ess.Mat. + D.Scale)

DPT-KITTI & SuperGlue (PnP)

DPT-KITTI & SIFT (Ess.Mat. + D.Scale)

DPT-KITTI & SIFT (PnP)

RPR [R(6D) +t]

RPR [3D-3D]

RPR [R(α, β, γ) + s · t(θ, φ)]

Method	AUC (VCRE < 90px)	Precision (VCRE < 90px)	Median Reproj. Error (px)	AUC (Err < 25cm, 5°)	Precision (Err < 25cm, 5°)	Median Trans. Error (m)	Median Rot. Error (°)	Estimates for % of frames
DPT-KITTI & RoMa (Ess.Mat. + D.Scale) Note: maybe buggy	0.669	45.6%	128.8	0.407	22.8%	1.23	11.1	99.9
KBR & LoFTR (PnP)	0.631	34.0%	167.5	0.277	10.5%	2.32	39.5	100.0
DPT-KITTI & LoFTR (PnP)	0.618	33.4%	166.7	0.269	9.8%	2.31	39.4	100.0
DPT-KITTI & LoFTR (Ess.Mat. + D.Scale)	0.614	34.7%	167.6	0.346	15.4%	1.98	30.5	100.0
DPT-KITTI & SuperGlue (Ess.Mat. + D.Scale)	0.602	36.1%	160.3	0.346	16.8%	1.88	25.4	100.0
DPT-KITTI & SuperGlue (PnP)	0.598	36.0%	156.9	0.252	10.7%	2.10	32.5	100.0
DPT-KITTI & SIFT (Ess.Mat. + D.Scale)	0.504	25.0%	222.8	0.252	10.3%	2.93	61.4	100.0
DPT-KITTI & SIFT (PnP)	0.468	25.1%	192.0	0.190	7.8%	3.35	63.7	100.0
RPR [R(6D) +t]	0.402	40.2%	147.1	0.060	6.0%	1.68	22.5	100.0
RPR [3D-3D]	0.387	38.7%	148.7	0.060	6.0%	1.69	22.9	100.0
RPR [R(α, β, γ) + s · t(θ, φ)]	0.354	35.4%	166.3	0.105	10.5%	1.83	23.2	100.0

Evaluation Leaderboard