Multi-Layer Depth Estimation

---

Task

Under this setting, we evaluate methods' multi-layer depth estimation on full LayeredDepth benchmark.

Metrics

Pair-wise, triplet-wise, and quad-wise accuracy on odd layers (layer1, layer3, layer5, layer7) of LayeredDepth benchmark. Refer to the paper for detailed definitions.

Leaderboard

	All			Mixed			Layer1			Layer3			Layer5
Method	P↑	T↑	Q↑	P↑	T↑	Q↑	P↑	T↑	Q↑	P↑	T↑	Q↑	P↑	T↑	Q↑
Multi-head (DAv2) [1]	82.56	70.57	61.34	90.41	71.91	60.23	85.81	74.41	65.62	72.87	60.75	54.36	63.99	56.42	51.99
Multi-head (NeWCRFs) [2]	63.95	41.65	25.32	74.91	46.22	25.44	65.59	39.85	24.16	54.22	38.39	27.6	47.33	34.28	27.11
Index Concat (NeWCRFs) [3]	66.35	45.25	26.9	80.74	52.45	28.67	66.72	41.42	24.05	55.66	41.61	30.01	52.26	40.34	29.86
Recurrent (NeWCRFs) [4]	62.26	40.7	23.77	69.4	42.01	22.97	68.13	44.64	25.91	48.59	30.42	20.5	42.41	27.73	20.34

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

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

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

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