[benchmark] evaluate the detectors on the AutoShot dataset

[awkrail]

Related to #484. In addition to BBC, I implemented benchmarking codes on the AutoShot dataset. The original work uses 200 videos for the test set, yet 36 videos are missing (link). Hence, I evaluated the codes on the remaining videos.

Results
The following table shows the results, indicating that AdaptiveDetector achieves the highest scores. This is consistent with BBC.

Detector	Recall	Precision	F1	Elapsed time (second)
AdaptiveDetector	70.77	77.65	74.05	1.23
ContentDetector	63.67	76.40	69.46	1.21
HashDetector	56.66	76.35	65.05	1.16
HistogramDetector	63.36	53.34	57.92	1.23
ThresholdDetector	0.75	38.64	1.47	1.24

[Breakthrough]

Awesome, thanks for the PR! Judging by the results for ThresholdDetector this dataset probably includes some more variety of transitions like dissolves/fades.

[awkrail]

@Breakthrough Yes, and I think that video cut type annotations (e.g., abrupt, dissolves, fade-in/out) may be useful. I will attach it with video cuts in the future for detailed evaluation.