1. What have the authors contributed in "Human-assisted motion annotation" ?
The authors propose a human-inloop methodology to create a ground-truth motion database for the videos taken with ordinary cameras in both indoor and outdoor scenes, using the fact that human beings are experts at segmenting objects and inspecting the match between two frames.
read more
![Figure 4. For the RubberWhale sequence in [5], we labeled 20 layers in (b) and obtained the annotated motion in (c). The “ground-truth” motion from [5] is shown in (d). The error between (c) and (d) is 3.21◦ in AAE and 0.104 in AEP, excluding the outliers (black dots) in (d). (e): The color encoding scheme for flow visualization [5].](/figures/figure-4-for-the-rubberwhale-sequence-in-5-we-labeled-20-1ijd6vng.webp)
![Figure 6. Some frames of the ground-truth motion database we created. We obtained ground-truth flow fields that are consistent with object boundaries, as shown in column (3), the horizontal component of flow, and column (4), flow colorization using Figure 4 (f). In comparison, the output of an optical flow algorithm [8] is shown in column (5). The error between the ground-truth motion (4) and flow estimation (5) is as follows (AAE, AEP), (a): 8.996◦ , 0.976; (b): 58.904◦ , 4.181; (c): 2.573◦ , 0.456; (d): 5.313◦ , 0.346; (e) 1.924◦ , 0.085; (f): 5.689◦ , 0.196; (g): 5.2431◦ , 0.3853; and (h): 13.306◦ , 1.567. Most of the errors are significantly larger than the errors with the Yosemite sequence (AAE 1.723◦ , AEP0.071) . The parameter of the flow algorithm in column (5) is tuned to generate the best result for each sequence.](/figures/figure-6-some-frames-of-the-ground-truth-motion-database-we-3nmnwowu.webp)