These are chat archives for soruly/whatanime.ga

18th
Sep 2016
soruly
@soruly
Sep 18 2016 07:42
@spillerrec currently the search endpoint has anti-spam by reCaptcha, there's no way to submit search requests in batch. This is because the server can only handle about 10 concurrent search requests. I'm developing an official API for that, but still there'd be rate limits. In such scale, you'd better off running this locally. Setting up the whole system is super complex though.
I'm still deciding whether to write tutorials to describe how to setup my exiting system, or should I rewrite part of my system so it'd be easier for other's to setup
soruly
@soruly
Sep 18 2016 07:51
Currently the whole system composes of PHP, javascript, shell scripts, python, elasticsearch, solr, nginx, nodejs and ffmpeg, you know how hard it is to setup everything to work
soruly
@soruly
Sep 18 2016 07:59
Oh, one more issue to concern:
If your is screenshot is taken from like OPs/EDs the system can't tell the real episode your screenshot is taken, because very likely the scene is exactly the same.
Sebastian Wahl
@spillerrec
Sep 18 2016 17:53
Even if you end up making the system easier to setup, you would still have to document it somewhat if you hope others can replicate it.
Sebastian Wahl
@spillerrec
Sep 18 2016 18:01
But what is done differently compared to basic LIRE (other than it being online)? As far as I could tell, the video are converted to a bunch of screenshots taken at even intervals, an image descriptor was chosen (why this one in particular?), and then it is offhanded to LIRE running on Solr. Any other things going on which I missed? How does elasticsearch come into the picture?
It seems like it might be easier just to try write a program using LIRE directly, than trying to set up this system intended for online use?