First Layer Depth

---

Task

Under this setting, we only evaluate methods' depth estimations for LayeredDepth's first layer relative depth tuples. For transparent objects, methods must predict the depth of the transparent occluder rather than the background.

Metrics

Pair-wise, triplet-wise, and quadruplet-wise accuracy.

Leaderboard

Method	Pair-wise accuracy	Triplet-wise accuracy	Quadruplet-wise accuracy
Metric3D V2 ft. [1]	89.53	81.71	75.2
DepthPro [2]	87.39	76.29	69.46
Depth Anything V2 [3]	85.34	74.44	70.43
Depth Anything [4]	78.02	62.95	58.88
Metric3D V2 [5]	80.31	65.43	55.14
MoGe [6]	76.76	63.99	58.92
MiDaS v3.1 [7]	76.61	62.05	58.54
Marigold [8]	82.59	68.35	55.89
GeoWizard [9]	81.39	66.29	52.43
UniDepth V2 [10]	77.03	62.15	56.86
ZoeDepth [11]	74.25	58.56	52.73

[1] Seeing and Seeing Through the Glass: Real and Synthetic Data for Multi-Layer Depth Estimation. [paper]

[2] Depth Pro: Sharp Monocular Metric Depth in Less Than a Second. [paper] [code]

[3] Depth Anything V2. [paper] [code]

[4] Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data. [paper] [code]

[5] Metric3D V2: A Versatile Monocular Geometric Foundation Model for Zero-shot Metric Depth. [paper] [code]

[6] MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training. [paper] [code]

[7] MiDaS v3.1 – A Model Zoo for Robust Monocular Relative Depth Estimation. [paper] [code]

[8] Repurposing Diffusion-Based Image Generators for Monocular Depth Estimation. [paper] [code]

[9] GeoWizard: Unleashing the Diffusion Priors for 3D Geometry Estimation from a Single Image. [paper] [code]

[10] UniDepthV2: Universal Monocular Metric Depth Estimation Made Simpler. [paper] [code]

[11] ZoeDepth: Combining relative and metric depth. [paper] [code]