ExplOCR: An OCR Application Explorers May Find Useful

Commander Wolzan

C
OamodSias said:
This is awesome, well done. :)

I was thinking of doing something similar a while back, a desktop app for explorers to capture the information of the systems they explore, store it in a DB and then upload it with images to share; only problem was finding a library that worked with my compiler, and which was O.S - unfortunately I was unlucky and gave it up as a bad job; now there's Elite Galaxy Online (the web version of the same idea, lol) and it's a beautiful looking site too. :)

Anyway, awesome someone wrote an OCR tool for this.
Click to expand...

Hey thanks commander :)
 

ThoroughlyLostExplorer

ThoroughlyLostExplorer
Hi Explorers,

I've just published the Beta 3 release, which you can find at https://github.com/ThoroughlyLostExp...plOCR/releases

Basically, you just need to download the ZIP archive for "Beta 3", unpack it and run from that folder.

However, I've just now noticed that this would let Beta1 or Beta 2 users start with an empty archive/database. You will have to copy the files manually. Fear not, the new Beta 3 version at last allows user-defined save file directories, so you will not have to do this again. ;) This time around, however, please follow the...

Upgrade Instructions for Beta1 and Beta2 Users

- Locate your old Beta1 or Beta 2 version ExplOCR folder. Rename it to ExplOCR.bak to keep it as a backup.
- Unpack the Beta2 ZIP archive and copy the folder inside to a location of your choice -- typically the same where you were running Beta1.
For simplicity, I'll assume the location is called ExplOCR in the instructions below.
- Locate your existing screenshot directory in the backup: ExplOCR.bak\OCR\screenshots and copy the complete directory to the OCR folder of your new installation to become e.g. ExplOCR\OCR\screenshots
- Locate your database file in the backup: ExplOCR.bak\OCR\save\systems.xml and copy it to the OCR folder of your new installation to become e.g. ExplOCR\OCR\save\systems.xml
- Do not copy the other xml files in the folder, those aren't user data.
- Start the ExplOCR application and open the table view by menu or the inviting green "Browse Archive" button.
- Your previous scans should be there.

As with previous releases, I'll add a list of new features etc. to the original post at the top of this thread. Let me know should anything not work as expected, remember this is a beta.

From what I have seen, recognition accuracy is currently quite good, and I'd like to ask you to to raise the quality standard you expect from it. I have noticed that the Beta 2 build had the habit of specifically failing to recognize the "EARTH MASSES" label (letter "H" likes to disconnect into two letters "I"). Ooops, sorry. I'll be looking to fix this type of issue quicker in Beta 3, should anything be reported to me. To get your "Earth Masses" back if it got removed from any scans, create a safety backup of your systems.xml database, then uncheck the "Read Only" box in table view and use the Re-Read button to process the archive copy of the scan. This should now detect that label properly.

Please be sure to send me some feedback. This can be bug reports or suggestions. I rely on this feedback to crerate a useful application: My own preferences in what an application should be like are somewhat idiosyncratic, so I need to hear from you to know how to make it useful for other people, too.

Best regards,

TLE

PS: My next step will be support for other screen resultions than 1920x1080 and 1920x1200. I have some ideas that will allow me to do this without spending too much time. However, I again need feedback e.g. on what resolution you view your system map in. Send me some feedback, please.
 

Genar-Hofoen

Genar-Hofoen
Greetings ThoroughlyLostExplorer!

I haven't had the time (or the energy) these last few weeks to do much with Captain's Log. However, I've been giving OCR a bit of thought recently, and the fantastic work you've been doing here does in fact seem like it would be something CL could utilise.

I haven't had the time yet to look at ExplOCR, but I'm wondering if there can be some way of running that and CL in parallel, with ExplOCR feeding information to CL.

ExplOCR is written in C#, but CL is written in Python, so the only way I can see of integrating the two would be some sort of API from which CL could interrogate the information garnered by ExplOCR. A common API would also be great for the excellent EDDiscovery and any other apps that might benefit from what ExplOCR provides.

When I get the time, I'll have a good look at it, but it looks like I'm going to be tied up for a while yet in other things.

In the meantime, keep up this awesomeness! o7

Regards

Genar.

P.S. I run ED usually in 1920x1080. Out exploring I usually run in Solo in order to get hi-res shots as well.
 
Last edited:

cygnusa

cygnusa
+ rep very interesting... I started exploring longer distances recently. I'm too time limited to log each unique planet/system visited by hand, and screenshots become meaningless (can't identify locations) if I don't closely document everything along the way. Something like this will help make the trip a bit more interesting... track progress and quickly snag details for future reference. Quite cool. I'm not super active due to limited time, but I log in a few days a week for a few hours each session.. so as such, taking notes isn't top on my to-do list. This solves a problem I had. I'll give it a try as soon as I can,
 

wolverine2710

Tutorial & Guide Writer
wolverine2710
Brilliant work, rep +1. And another gem I missed when searching for entries in EDCodex. The admin tools of EDCodex showed me you just added it yourself. Highly appreciated. Very nice to see that more and more of the ED screens gets succesfully OCR-ed.

EDCodex is a website with a database of currently approx 210 tools,threads,websites,videos for ED. Any one can and is encouraged to add entries there. EDCodex companion thread. Its equally suited for PC's, tablets and smartphones.

Note: I'm going to mention you and Elite Galaxy Online by Commander Wolzan in the EDCodex thread. You basically around the same time got OCR-ing working for the ED system view. The NEXT step in OCR-ing ;-) Brilliant work chaps. Now that the gauntlet has been thrown I'm sure more will follow ;-)
 
Last edited:

ThoroughlyLostExplorer

ThoroughlyLostExplorer
Hi Explorers,

I just noticed a number of failures in OCR reading that weren't there before. These result in lines of "?????" within the text display that should instead contain text. Of course, those lines will also be missing from the grid view.

As it turns out, there are detail differences in the way letters are rendered between screenshots I made a few weeks ago and screenshots made now. Unfortunately, it seems that the systems I scanned days ago currently show up as unexplored to me so I can't compare screenshots for those without flying there. I'm somehow inclined to believe the letter rendering differences have something to do with 1.4.

The differences I observe aren't too large, but they will prevent the current neural net from identifying some letters in some cases. Unfortunately, the differences can also cause two neighbouring letters to "stick together", which causes neither of them to be identified. For this reason, words like "AXIS" (X sticks to A), "TYPE" (T sticks to Y) or "METAL" (E sticks to T) have become hard to read. Obviously, this creates problems for data table items containing those words.

I haven't had a time to look at a lot of OCR, but number recognition so far seems fine. It is however only a matter of time before someone encounters a case where numbers are not read well because I suspect the minus sign in negative numbers will use any chance it can get to stick to the adjacent digit.

To fix these issues, I will have to retrain all neural networks involved in the OCR process. I'd have to do so based on new screenshots, and selecting these may have to be based on different criteria as previously to ensure good recognition quality. Also, I will have to find filtering parameters which keep letters from sticking together but which don't cause other kinds of trouble.

At the moment, I don't have the time to do this. I'll create an improved version with adjusted filters but based on thre old neural nets, which should fix most of the problems, in the next few days. Until then, please be patient.

TLE

PS: Actually, the "new" letters on screenshots seem to be slightly more suited to OCR than the "old" ones. Once I have enough time to train the neural nets, we may end up with better recognition quality than before.
 

ThoroughlyLostExplorer

ThoroughlyLostExplorer
Hi Explorers,

it's me again. I've decided to rush out a hotfix that will fix the most urgent problems since game version 1.4. The new version should be equivalent to Beta 3 in most aspects. Download address is https://github.com/ThoroughlyLostEx...nload/TagBeta3-1.4-a/ExplOCRbeta3-for-1.4.zip

Installation should be be straightforward (unpack into directory of choice) because in Beta 3, user data is located in a custom path that ideally is outside the install folder. To be on the safe side, make a backup of your uder files. Check the "Config" / "Save Directories" dialog for your current file location.

I doubt that I've seen all issues caused by the slightly different typeface rendering in version 1.4 (... if that even is what is causing them), hence some issues may persist despite the change. I'll give the new version a few more screens to OCR, but unfortunately I have very little time at the moment.

TLE
 

Satsuma

Satsuma
This is pretty neat, I'll definitely be giving this a shot. I definitely agree with Finwen and Kancro, it would be fantastic if this worked with EDDiscovery, or Captain's Log...
 

ThoroughlyLostExplorer

ThoroughlyLostExplorer
Hi Explorers,
I haven't had a look at the program in the Horizons or 1.5 betas because I'm not a beta user. Actually, I haven't had a look at it in quite a while because I've been rather busy. I'll wait until the release is available and playable to see what happens. The 1.4 patch actually did cause some trouble, even though the font typeface rendering changes were barely visible to the eye. At this point, I wouldn't know even if they changed the font completely.

If something does happen, I'll see if I have enough time available for a hotfix before christmas. Otherwise, I'll make a fix soon next year.

Best,

TLE

PS: I didn't actually get much feedback for the 1.4 hotfix, I'm however assuming that it worked because I haven't been getting complaints either.
 

ThoroughlyLostExplorer

ThoroughlyLostExplorer
Hi Explorers,

I've downloaded Horizons now and as it turns out they have changed the font size quite a bit, plus added some new items my program can't cope with very well. aslo, degrees are now '°' instead of 'DEG'. All of this can be fixed, but I'll have to retrain the software on a new set of sample screenshots. This will take some time, and I will probably not get around to doing that before the holidays.

The program will still generate reasonable output for many lines. Others will goof up, like 'AXIAL TILT' being read as 'RADIUS' because of the font size change. I'm rather keen on producing dependable quality of recognoition. So I consider the program to be not usable at the moment.

On the positive side, I am confident that the new font will eventually produce good results after retraining because the typeface looks quite 'clean' on my monitor and not as fuzzy as it used to.

Of course, I'm happy to have Horizons available, even though it will mean a bit of work for me. I'll spend 100x as much driving around on planets, of course.

Have fun landing/driving, and please be patient until I get around to a fix in the new year.

TLE

Addendum: Things look a little less grim today since I remebered that it is possible to repair "bad" OCR data the program generates now after the fix in January.
 
Last edited:

ThoroughlyLostExplorer

ThoroughlyLostExplorer
Thanks guys,

I really appreciate your positive feedback -- but please keep my post above in mind. Since 1.5/Horizons was released, the application will mis-read parts of the planet data because the font was changed and some content was added. Currently, it really doesn't offer the recognition quality an application of this type should have, and it needs to be fixed before you can really use it.

Before the holidays, I definitely don't have the time to do all the exploring I'd need to do to be reasonably sure I've seen all problems (and new features!) that result from the new version of the game. I'm sure a fix that really works is much better than a quick fix that works only part of the time.

There is, however, a bit of a workaround you can use if you do want to go exploring and don't want to miss out on documenting your data. The most important part of this is that the application will automatically create an archive of screenshots from OCRed systems, and these can in principle be OCRed again once a fix is in place. The tricky thing is to keep track of the system name, because those are quite hard to OCR.

The following should work:


  • Make a note of where 1.5 came along for use in fixing data later. The numbers in column "ARCHIVE_NAME" in menu Table / Display are useful here.
  • Use the program as you nomally would, saving your OCRed data as you go.
  • Be aware that the data in menu Table / Display will contain bad data.
  • The program will store system name and the image archive automatically.
  • After the fix comes along (-> January), sort the table by ARCHIVE_NAME and use the "Re-Read" feature for all systems scanned since then. This will change the data values, but preserve the system name and administrational metadata

I'm afraid re-reading will be a bit of a hassle if you have scanned many systems. I am confident however that I can easily automate the feature to re-read many systems at once, so all you'd have to do is make some tea while your PC happily re-reads the OCR.

Thank you for your patience,

TLE
 

Jon474

Banned
Jon474
TLE, your suggested process sounds like an excellent workaround. That will do for me!

Thank You

Kind regards
Jon
 

ThoroughlyLostExplorer

ThoroughlyLostExplorer
Hi Explorers,

in case any of you were wondering about the fix for 1.5/Horizons: I'm back from the holidays and will get to work on it. If no unforseen problems present themselves, may be done by monday. I've noticed a major expedition is planned, will try to be done in time (17. Jan?) for anyone who might want to use ExplOCR to keep a log.

Haven't done any exploring recently, has anyone noticed changes in the data presented by the game besides the new "gravity" item? New atmosphere or volcanism types, maybe even new types of stars?

TLE
 
You must log in or register to reply here.
Back
Top Bottom