EUROGRAPHICS 2012: SimpleFlow

Supplement for "SimpleFlow: A Non-iterative, Sublinear Optical Flow Algorithm"

Performance

EUROGRAPHICS 2012
Michael Tao, Jiamin Bai, Pushmeet Kohli, Sylvain Paris

Instructions: click desired image to view full-sized image.

HD Timing Performances

(4K: 4096 x 2304)

Pair 1

Frame 1

Frame 2

Our algorithm without sublinear optimization

Our Sublinear Scheme

128 x 72

256 x 144

512 x 288

1024 x 546

3048 x 1152

4096 x 2304

Timings

Our Sublinear Complexity:
~(N^0.779)
Our algorithm without sublinear optimization Complexity:
~(N^1.0007)

(1080P: 1920 x 1080 Resolution)

Pair 1	Frame 1	Frame 2	Our algorithm without sublinear optimization	Our Sublinear Scheme	480x270 optimization	960x540 optimization	1920x1080 optimization	Timings Our Sublinear Complexity: ~(N^0.53) Our algorithm without sublinear optimization Complexity: ~(N)

Pair 2	Frame 1	Frame 2	Our algorithm without sublinear optimization	Our Sublinear Scheme	480x270 optimization	960x540 optimization	1920x1080 optimization	Timings Our Sublinear Complexity: ~(N^0.92) Our algorithm without sublinear optimization Complexity: ~(N)

Pair 3	Frame 1	Frame 2	Our algorithm without sublinear optimization	Our Sublinear Scheme	480x270 optimization	960x540 optimization	1920x1080 optimization	Timings Our Sublinear Complexity: ~(N^0.46) Our algorithm without sublinear optimization Complexity: ~(N)

Pair 4	Frame 1	Frame 2	Our algorithm without sublinear optimization	Our Sublinear Scheme	480x270 optimization	960x540 optimization	1920x1080 optimization	Timings Our Sublinear Complexity: ~(N^0.48) Our algorithm without sublinear optimization Complexity: ~(N)

Pair 4	Frame 1	Frame 2	Our algorithm without sublinear optimization	Our Sublinear Scheme	480x270 optimization	960x540 optimization	1920x1080 optimization	Timings Our Sublinear Complexity: ~(N^0.32) Our algorithm without sublinear optimization Complexity: ~(N)

Middlebury Flow Quality Performances
Refer to Figure 5 for descriptions of each image.

Dimetrodon	Frame 1	Frame 2	Our algorithm without sublinear optimization (RMSE: .323)	Our algorithm with sublinear optmization (RMSE: .322)	Our algorithm followed by Sun et al. 2010 (RMSE: .165)	Sun et al. 2010 only (RMSE: .166)	ground truth
Grove 2	Frame 1	Frame 2	Our algorithm without sublinear optimization (RMSE: .580)	Our algorithm with sublinear optmization (RMSE: .579)	Our algorithm followed by Sun et al. 2010 (RMSE: .316)	Sun et al. 2010 only (RMSE: .318)	ground truth
Grove 3	Frame 1	Frame 2	Our algorithm without sublinear optimization (RMSE: 1.458)	Our algorithm with sublinear optmization (RMSE: 1.460)	Our algorithm followed by Sun et al. 2010 (RMSE: 1.372)	Sun et al. 2010 only (RMSE: 1.398)	ground truth
Hydrangea	Frame 1	Frame 2	Our algorithm without sublinear optimization (RMSE: .539)	Our algorithm with sublinear optmization (RMSE: .539)	Our algorithm followed by Sun et al. 2010 (RMSE: .399)	Sun et al. 2010 only (RMSE: .398)	ground truth
Urban 2	Frame 1	Frame 2	Our algorithm without sublinear optimization (RMSE: 1.436)	Our algorithm with sublinear optmization (RMSE: 1.464)	Our algorithm followed by Sun et al. 2010 (RMSE: .854)	Sun et al. 2010 only (RMSE: .804)	ground truth
Urban 3	Frame 1	Frame 2	Our algorithm without sublinear optimization (RMSE: 2.064)	Our algorithm with sublinear optmization (RMSE: 2.070)	Our algorithm followed by Sun et al. 2010 (RMSE: 1.487)	Sun et al. 2010 only (RMSE: 1.166)	ground truth
Venus	Frame 1	Frame 2	Our algorithm without sublinear optimization (RMSE: .568)	Our algorithm with sublinear optmization (RMSE: .568)	Our algorithm followed by Sun et al. 2010 (RMSE: .429)	Sun et al. 2010 only (RMSE: .426)	ground truth
Whale	Frame 1	Frame 2	Our algorithm without sublinear optimization (RMSE: .383)	Our algorithm with sublinear optmization (RMSE: .382)	Our algorithm followed by Sun et al. 2010 (RMSE: .271)	Sun et al. 2010 only (RMSE: .272)	ground truth