Motion estimation benchmark
Results
In the following we divided the results into two tables. The static evaluation table groups results using only the final result of algorithms, whereas the dynamic evaluation table contails results on the full dynamics. This distinction is necessary since a lot of the classical approaches where not made with dynamics in mind, and thus lead to very poor results.
Each table lists the score obtained by each of the considered approaches on each stimuli. This score is always in the [0,1] range. The scoring methodology we set up for each experiment is linked into the table headers. A dash (—) instead of a score indicates that the experiment has not been made, whereas a N/A denotes a non applicability of the experiment on a given approach.
Colour palette:
0.0 (worse)
1.0 (better)
Static evaluation
| Average | Grating size | Translating diamonds | Barberpole | Translating dashed bar | |
|---|---|---|---|---|---|
| tlapale-masson-etal:10 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
| sun-roth-etal:10 | 0.86 | 0.65 | 1.00 | 0.78 | 1.00 |
| brox-malik:10 | 0.74 | 0.00 | 1.00 | 0.98 | 1.00 |
| bayerl-neumann:07 | 0.68 | 0.36 | 1.00 | 0.38 | 1.00 |
| lucas-kanade:81 | 0.45 | 0.99 | 0.00 | 0.00 | 0.81 |
| horn-schunck:81 | 0.39 | 1.00 | 0.00 | 0.03 | 0.52 |
| BMOCV | 0.19 | 0.44 | 0.32 | 0.00 | 0.00 |
Dynamic evaluation
| Average | Grating size | Translating diamonds | Barberpole | Translating dashed bar | |
|---|---|---|---|---|---|
| brox-malik:10 | — | 0.18 | 0.36 | — | 0.50 |
| sun-roth-etal:10 | — | 0.00 | 0.36 | — | 0.75 |
| horn-schunck:81 | — | 0.00 | 0.03 | — | 0.75 |
| BMOCV | — | 0.05 | 0.36 | — | 0.50 |
| lucas-kanade:81 | — | 0.00 | 0.36 | — | 0.75 |
| tlapale-masson-etal:10 | — | 0.08 | 0.96 | — | 1.00 |
| bayerl-neumann:07 | — | 0.07 | 0.39 | — | 0.50 |