Multi-Layer Optical Flow

---

Task

Under this setting, we evaluate methods' multi-layer optical flow predictions on full LayeredFlow benchmark.

Metrics

• Multi-Layer Bad-τ denotes the percentage of pixels with flow prediction that is not τ-accurate
• Multi-Layer Count-Aware Bad-τ denotes the percentage of pixels with either incorrect layer count or flow prediction that is not τ-accurate. We use τ=1, 3, and 5 pixels.

Refer to the paper for detailed definitions.

Leaderboard

	Layer1				Layer2				Layer3
Method	1px↓	3px↓	5px↓	∞px↓	1px↓	3px↓	5px↓	∞px↓	1px↓	3px↓	5px↓	∞px↓
MultiRAFT (L) [1]	76.52	51.84	42.61	9.19	91.72	79.92	73.46	47.2	98.0	88.12	86.75	39.0
MultiRAFT (S+L) [2]	77.81	54.93	45.42	8.85	88.76	75.04	63.77	40.76	94.25	85.62	83.0	21.25

[1] LayeredFlow: A Real-World Benchmark for Non-Lambertian Multi-Layer Optical Flow. [paper] [code]

[2] LayeredFlow: A Real-World Benchmark for Non-Lambertian Multi-Layer Optical Flow. [paper] [code]