Create test videos automatically

[TimothyGillespie]

Videos for testing should be auto generated to monitore accuracy and edge cases. They might not cover some naturally occuring cases which shall not be their scope.

[Korosensei42]

For the input I'd recommend

• a minimal brightness (int between 0 and 255)
• a maximal brightness (int between 0 and 255)
• a time period (int)
• a video length (int)

I imagine the output to be a video that has two monochrome grayscale images, a darker and a brighter one. It changes back and forth between these two images after each time period.

Is this the direction you were going for?

[TimothyGillespie]

I didn't have this idea, but I like this too. My idea was to have something auto-generate like a box floating from one position to another (this kind of thing caused a false positive). Or a flashing section in only part of the video (which might be realized with your idea actually).

So I had even imagined more input options on top of yours. My general idea being to try tricking the detector. The two gray scale images would work really well to test the general measurement and detection, but they can be detected with a naive approach.

I will definitely include this and would like to hear if you see any flaw in my approach / base idea.

[TimothyGillespie]

I forgot to mention. I plan to put this into it's own repository as well. Though maybe I am keeping things too modular with this kind of approach.

[Korosensei42]

Maybe I don't understand your floating-box-idea. However, from what I understand it could look like this:

You have a black background with a red box in the middle of it. In the next frame the box disappears and appears at some other position again. repeat this process at an arbitrary frequency arbitrarily often and that is your test video. Is this what you mean?
If so, I can definetly imagine that you get false positives.

The other idea you brought up features the same thing, only that the box now stays at the same position. Did I understand this correctly?

In general I don't see any problems with your approaches. We could break all ideas up to this point down like this:

You want to quickly change the following characteristics of an object on screen:

• the position
• the brightness
• the frequency in which these changes take place
• the color (maybe ?)
• maybe you also want to think about the form of the object (i.e. changing a square to a circle from one frame to another)

For the repo, I'd think either one would be fine. I would recommend going with the option that fits your working-style the best.