- Dec 29, 2017
- 3,209
- 5,574
Btw while working on the OCRSMTL stuff I learned a lot of new things and I'm once again revamping my translations.
(No don't worry I'm not gonna give it a new fancy name, I'm just gonna stick with calling everything SLR lol)
I have been using Reddo9999's code for the initial Sugoi input for ages, because I didn't want to (and in parts couldn't) write my own. (That's what the R in my stuff stands for.)
But now that I actually understand programming more, and how AI functions, I've started to re-write it, and it turns out there's actually a huge room for improvement here. I had my SLR system kinda backwards, in that I had my main focus on fixing errors of the Sugoi output instead of modifying the input in a way that the errors don't occur in the first place.
As with all fun experiments of mine that means the next translations will be much more prone to crashing and weirdness, but if this works out I might be able to drastically reduce the time it takes to make a SLRMTL even further, while actually (slightly) improving translation quality overall.
First tests look really promising (Much faster.), but the spooky bit is that this for now also gets rid of a bunch of fail-safes I had. So this could end in disaster.
Just as general information for the next poor soul requesting something.
Edit: And just to be stupid about it, SLR will now cover every single not manually edited translation from me and now means:
Shisaye (So it can still be used if something better than Sugoi comes along.)
L8rdude (For the groundwork of the picture based text translations.)
Reddo9999 (For the groundwork of the feeding system.)
(But ofc it's also still the "Self Loading Rifle" pun.)
(No don't worry I'm not gonna give it a new fancy name, I'm just gonna stick with calling everything SLR lol)
I have been using Reddo9999's code for the initial Sugoi input for ages, because I didn't want to (and in parts couldn't) write my own. (That's what the R in my stuff stands for.)
But now that I actually understand programming more, and how AI functions, I've started to re-write it, and it turns out there's actually a huge room for improvement here. I had my SLR system kinda backwards, in that I had my main focus on fixing errors of the Sugoi output instead of modifying the input in a way that the errors don't occur in the first place.
As with all fun experiments of mine that means the next translations will be much more prone to crashing and weirdness, but if this works out I might be able to drastically reduce the time it takes to make a SLRMTL even further, while actually (slightly) improving translation quality overall.
First tests look really promising (Much faster.), but the spooky bit is that this for now also gets rid of a bunch of fail-safes I had. So this could end in disaster.
Just as general information for the next poor soul requesting something.
Edit: And just to be stupid about it, SLR will now cover every single not manually edited translation from me and now means:
Shisaye (So it can still be used if something better than Sugoi comes along.)
L8rdude (For the groundwork of the picture based text translations.)
Reddo9999 (For the groundwork of the feeding system.)
(But ofc it's also still the "Self Loading Rifle" pun.)
Last edited: