codensuch

ChatGPT's Translation Skills are So Good, You'll Think a Human Did It!

I've added ChatGPT in my development build. Look at the results above. You tell me.

And yes, I asked ChatGPT to come up with the title :P

Edit: Uploaded image was wrong size. Should be fixed now.

Demystifying common translation issues

I thought it might be fun to address some of the common "issues" that you may see during the alpha tests. I am aware of almost all issues, but may not address them in a timely manner. Hopefully this post demystifies how I tackle these issues based on my development style.

Inconsistent font size in bubbles

Background: Professional publications will tend to stick with a preset font sizing scheme regardless of original Japanese font size. They tend...

A whole new AI world

In my last post, I mentioned that I set out 4 upcoming milestone. First of which was an UI overhaul that puts the emphasis on the human translator. Over the last few weeks I ended doing my milestone 2 first...

Milestone 2: New machine learning models

I was on a roll training new machine learning models and decided to just see it through and hold off on the editor UI. I experimented with brand new models that did some amazing things. The two that I want to show o...

Will be working extensively on the app over the next 3 weeks. Stay tuned.

Hoping to have the new ML models deployed by then. Cheers!

Version 2

I apologize for the slow updates. There are stuff happening IRL that I just can't get away from, but I am working on the app! I've been doing a bunch of experiments on getting the app to the next level, and I believe I figured out all the missing pieces to solve fully automated translation. This includes automated translation of ALL text (not just those inside speech bubbles). Let's call it version 2. I currently have 4 major milestones. The idea is to get something working o...

Life has been weird and busy... Human assisted cleaning coming

Sorry for the lack of updates. As title indicates, life was in the way. But don't worry, I'm working on it whenever I can get the time.

Here's an early look on something I'm working on. I'll have more to say later.

Improving OCR and continued public alpha test

I've been trying to improve the accuracy of OCR over the last couple of weeks. The new OCR AI is currently up on the public alpha site for you to check out. It is a culmination of a few things:

1. New OCR LSTM model. The previous model was using a modified version of the Tesseract's best Japanese model. This one is trained from ~30,000 lines of text from OPM manga. It took awhile to train a model that is somewhat on par with default model. It handles things like Kanji and punctuation mu...

I'm not limiting bubble count for Friday's alpha test!

Public alpha test #2 coming Friday 12pm PT! The test will last until Sunday midnight and I'll be monitoring the server throughout.

This test features fully automated speech bubble clean, translation, and typesetting. Like last time the primary goal is to test the stability of the server with these features turned on. I mentioned in my last post that I was going to limit the number of translated bubbles. I've changed my mind. All detected bubbles will be translated and typeset. However, ...

Public alpha testing #2 - now with translation - April 29th 12pm PT - May 1st 11:59pm PT

I'm happy to announce the second alpha test is coming on April 29th! This time we're testing the stability of translation/typesetting AI portion of the server. Since I'll be enabling the translation AI, this test will run in limited manner to prevent abuse. Therefore, all pages will be:

• Watermarked
• Typeset with Arial font only
• Half translated (e.g. 5 out of 10 speech bubbles will be translated, remaining will be cleaned)

The test will be much like the...

AI server has been up for over 2 weeks, 24/7! Next steps

Thanks to everyone who participated in the first ever public alpha stability test of my AI server! As a result I was able to fix a few bugs that caused disconnects and server crashes. The server has been live for the past two weeks with no disconnects and crashes so far. I'm going to keep it live 24/7 until further notice. Please feel free to continue to run tests on it.

Next alpha test is coming

I'm currently planning a second alpha test that involves translation...

Public alpha testing this weekend, link inside

I'm running the first ever public alpha test of OneTranslate AI server this weekend Feb 19-21. I'll be monitoring the server the entire weekend and fix any issue that arise during the test. Please report all issues and feedback in this the public-alpha-testing Discord channel.

This is a first test for server stability so only bubble cleaning feature is enabled.

First ever public alpha testing on Feb 19th 12pm PST to Feb 21st 11:59am PST

The first ever public alpha testing of OneTranslate is coming! The primary goal is to test the stability of the servers. Being a first test, only the bubble cleaning feature provided. It will be done over the 3 day long weekend so mark your calendars. If you wanted do some automated bubble cleaning this is the time!

I expect we only need about 20 concurrent users to stress the server enough to produce useable results. Thus, I will only advertise on the Patreon + OneTranslate Discord cha...

Dedicated AI server build complete! AI performance improvements incoming

First of all, happy new year! Hope everyone had a great holiday. I got a good chunk of time to work on the project and I have some great progress to report!

AI dedicated server is done!

The dedicated AI server well and working! All of the software has been setup and migrated over to the dedicated server. Future changes can also be quickly deployed to the dedicated server. The server will go live soon in the future.

Performance speed-up

Insight into overlay text segmentation, and dedicated server update

I'll get the latter out of the way first. The dedicated server is about half way built. I just got Tensorflow built on it and is in process of building and hooking up rest of software. Vacation time is coming up so I'll be able to dedicate way more time to work on this project. I expect to have something working before end of the year.

In other news, I've been experimenting with overlay text segmentation. It recently came up frequently as a topic of conversation with several Patreon mem...

Batch mode done! What's coming next will be a big deal…

Batch mode is finally done! You'll now be able to upload multiple pages, clean, translate, and typeset all in one place. There are two buttons that allows to download all edited or cleared pages.

This marks a significant milestone in the development of the app as the current feature set is complete enough to be actually useful in production. It is also important to understand that this app is still a prototype. So the next step in the development process is to validate the concept publi...

Quick batch mode update

I've been especially busy the past month so progress is a bit slower than usual. However, batch mode is coming along nicely! All of the core features are in place and most of the remaining work is about hooking up the UI. Soon, you'll be able to batch clear, translate, edit and/or download all in one place.

Stay tuned!

Work in progress update on an even better batch mode

The original plan was add a batch mode that allows the user to upload multiple pages simultaneously and clean all the bubbles. This feature is probably the most asked for as bubble cleaning is a very time consuming task for a human.

After a few design iterations of how this would work, I came to the conclusion that the incremental work required to get a full multi-page editor working along with batch is actually very little. So now I'm working on a full batch clean and editing feature t...

Server side rendering! A comparison

If you followed my last post you'll remember that one of the reported issue was that the font rendering on Windows looked too thin while macOS and Linux doesn't have this issue. ClearType on Windows seems to affect text rendering regardless of which browser was used. So in order to produce consistent text rendering across all platforms, all text are now pre-rendered on the server side using a custom renderer. Then the image is sent to the web app and applied over the page. There are other a f...

Quality of life updates (WIP)

This one is all about making the app simpler to use. All of changes are based off user feedback.

Font size selection changed to + - buttons

The automated line breaking doesn't guarantee that the selected font size will work. For example, the algorithm determines that 17pt is the best font size for this bubble using the current text, but you want something smaller. You try font 16pt and 15pt, both works but 14pt somehow doesn't. But oh, 13pt works… Even worse, 14...

Upgrading machine learning hardware!

I started this project by training on a GTX 970M, yeah a mobile graphic card. It was slow as hell and training took days for mediocre results. But eventually I got down to a model that gave an usable accuracy and I decided to go all out with a GTX 1080Ti (I've been using AMD on desktop for years). This was right after crypto crash in 2018 so the used market was flooded with cheap cards so I managed to get a used 1080Ti for cheap. The models that I use today only required overnight to train, s...

Bold and italic done! (maybe underline in the future)

This one was more tedious than expected, largely because the rendering needs to backend and frontend to match in order to produce accurate line breaking. So sets of rendering logic need to be written and synced on completely different platforms, and all rendering implementations are custom. On top of that I needed to build a mini text editor to the web client that can handle live formatting changes.

Underline also turned out to be something not built into browser canvas text rendering, ...

Week 1 Update

I had to the privilege of chatting with a few experienced translators and typesetters and got some feedback on current state of the app. Here are the top 3 backlog based on the aggregate feedback:

1. Bold, italic, underline

2. Quality of life changes (auto defaults, rendering issues, center alignment etc)

3. Batch cleaning mode

The first two are to address usability issues and editing feature completeness. Batch mode will allow the user to clean in batch which would ma...