Am 18.09.16 um 12:26 schrieb Lawrence D’Oliveiro:
On Sunday, September 18, 2016 at 10:13:41 PM UTC+12, Christian
Gollwitzer wrote:


Am 18.09.16 um 12:03 schrieb Lawrence D’Oliveiro:

Like I said, trying to automate a GUI is a waste of time. GUIs
are designed for humans, not computers, to use.

You don't always have a choice. Consider batch-processing a number
of images (say, 30,000 movie frames) using a proprietary effect
filter which works on single images only. It might be easier to
drive the GUI than to recreate the commercial algorithm.

Considering the power available in Free Software toolkits like
ImageMagick, G’MIC and so on, not to mention libraries accessible
from Python itself, let me suggest that such proprietary software
simply isn’t worth bothering with any more.

I was expecting that argument. Free software gives you a lot in this area, but there are commercial signal processing programs with unmatched quality. Examples:

https://ni.neatvideo.com/
 or for audio
http://www.celemony.com/en/melodyne/what-is-melodyne

Their business model relies on (nontrivial) secret algorithms. Even if the algorithm is published it might be too tedious to reimplement it from the scientific paper because it takes months of work or resources you don't have; for instance, this algorithm

http://graphics.cs.cmu.edu/projects/scene-completion/scene-completion.pdf

requires you to have millions of stock photo images, which you are allowed to use. If you are flickr, getty or Google, you have no problem. Otherwise it will be hard.


I agree with you that it is a shaky solution, but that doesn't make
it impossible.

Is it really something you want to entrust mission-critical business
functions to?

As always, it depends. I wouldn't rest my money on a bank account which is managed by driving Excel via Sikuli. OTOH, I wouldn't want to spend years of work to recreate NeatImage to denoise the photos of my last holiday trip. Here, driving the GUI is fine.

        Christian

--
https://mail.python.org/mailman/listinfo/python-list

Reply via email to