RegulatedNoise - trade tool with zxctypo's best-in-class OCR (EliteOCRReader as-was)
- Thread starter maxh2003
- Start date
I noticed this too. The automatic jumped to entries are not happening. Manually created entries are still ok.
Thanks for putting the work in, and most of all for open sourcing it.
Awesome tool. When it works for me its great. However getting a few fatal crashes. Dump file can be found.... here -> https://drive.google.com/file/d/0B3CFXabnEcvzSE5vMGVucS1vZ1k/view?usp=sharing
While I suspect that my problem is the same as those above I'll add my dump file to their's in the hope that it might help track down the problem.
The tool has started crashing as soon as it attempts to OCR a new screenshot. Happened out of the blue yesterday after successfully scanning several systems without problems.
Here's the dumpfile, I had to 7z it as it was pretty huge and I have really slow upload. http://1drv.ms/1un9UJ3
The tool has started crashing as soon as it attempts to OCR a new screenshot. Happened out of the blue yesterday after successfully scanning several systems without problems.
Here's the dumpfile, I had to 7z it as it was pretty huge and I have really slow upload. http://1drv.ms/1un9UJ3
Amazing tool.
I love RN. It works well for me even with a triple head display and a very wide FOV (using high-res screenshots w/autodelete).
I'm a little curious about the OCR. It's the only part of RN that stumbles. Is there no additional hinting after the initial recognition?
Often the OCR is good at reliably picking out one word of a two word commodity. For instance, it almost always gets 'fruit' right but mangles vegetables. Is there no second step of analysis in the OCR such that after the first pass the OCR says FRUIT ODE UEDTOODE or whatever, RN will suggest FRUITS AND VEGETABLES? I'm getting so tired of typing in the same handful of mis-recognized names that I'm getting ready to make a series of AHK autocomplete codes for them. It seems like given that the commodity list is a pretty small dictionary it would be possible to make a very strong guess in most situations, before giving up and displaying a raw text string from OCR, that we know isn't a commodity in the game.
I guess it would be a pain in the butt to support all the languages the client does with such extra hinting, but maybe it could be modularized (XML?) and the community could contribute to hint files for various languages.
I love RN. It works well for me even with a triple head display and a very wide FOV (using high-res screenshots w/autodelete).
I'm a little curious about the OCR. It's the only part of RN that stumbles. Is there no additional hinting after the initial recognition?
Often the OCR is good at reliably picking out one word of a two word commodity. For instance, it almost always gets 'fruit' right but mangles vegetables. Is there no second step of analysis in the OCR such that after the first pass the OCR says FRUIT ODE UEDTOODE or whatever, RN will suggest FRUITS AND VEGETABLES? I'm getting so tired of typing in the same handful of mis-recognized names that I'm getting ready to make a series of AHK autocomplete codes for them. It seems like given that the commodity list is a pretty small dictionary it would be possible to make a very strong guess in most situations, before giving up and displaying a raw text string from OCR, that we know isn't a commodity in the game.
I guess it would be a pain in the butt to support all the languages the client does with such extra hinting, but maybe it could be modularized (XML?) and the community could contribute to hint files for various languages.
Last edited:
Hi,
The tool is GREAT ! and still I am waiting for a donate button.. ;-)
My question:
When I start RN, the tool ask for an outgoing connection through my firewall. It works when i deny or accept this outgoing connection. But just to clear out any suspicion, what kind of information is collected outside. As I don't activate the box "Post data to EDDN on import".
Thanks for your help guys.
davidyannis
The tool is GREAT ! and still I am waiting for a donate button.. ;-)
My question:
When I start RN, the tool ask for an outgoing connection through my firewall. It works when i deny or accept this outgoing connection. But just to clear out any suspicion, what kind of information is collected outside. As I don't activate the box "Post data to EDDN on import".
Thanks for your help guys.
davidyannis
Hi, Since version 1.7 RN checks with GitHub for the latest version on startup, line 38 - https://github.com/stringandstickyt...mmit/4b6310d8333327d868f5b5afd7ab6f98f9988529
So nothing suspicious here, as you can see the source code for the whole project, other than RN making sure you are aware of the latest version.
Also if you have the webserver running that will need 'access' outside your PC to be seen on your local network devices.
Thanks
Last edited:
Good work on the updates but like I said in my last post the Commander's log is still not working properly. Now it's updating allright but it keeps updating the log even when I move inside the current system and not jumping to a new one. The log now shows Jumped to events every 10 seconds.
Last edited:
Minor issue - since update to 1.83 the OCR scan lists the system name as the 'previous' system visited and not the correct 'current' system. Although easy enough to amend, I hadnt realised for 4 system jumps and manually had to edit the systems so they were correct.
Is this issue unique to me or is anyone else experiencing the same?
EDIT: Rolled back to 1.81 to test and system name is scanning correctly.
2nd EDIT: It appears if you close and re-open the app (v1.83) before performing OCR it will pick up the current system correctly. I landed and scanned and it was wrong. Closed app, re-opened and re-scanned and it was correct.
Is this issue unique to me or is anyone else experiencing the same?
EDIT: Rolled back to 1.81 to test and system name is scanning correctly.
2nd EDIT: It appears if you close and re-open the app (v1.83) before performing OCR it will pick up the current system correctly. I landed and scanned and it was wrong. Closed app, re-opened and re-scanned and it was correct.
Last edited:
Hi all, have been using EliteOCR and heard that this is a better solution so I thought I'd give it a try.......and I'm stuck on step 1
Running 1.83 with ED running in fullscreen mode @ 2560x1440. Windows 8.1 x64
I can't perform a calibration. When I click on the Calibration tab all I get is a blank screen with ERROR in the resolution box. There are screenshots in my Pictures/Frontier Developments/Elite Dangerous folder.
Am I missing something here ? It Appears (according to the guide) that I click the Calibrate! button and it should ask me to browse for a screenshot to calibrate on.
As soon as I do this I get a fatal error message and the program closes.
Tried running as Admin, same result. Both the x86 & x64 Visual Studio Redistributables are installed.
Any ideas ?
The dump file is here http://goo.gl/NfSvQC (I RAR'd it as it was huge)
Cheers
Running 1.83 with ED running in fullscreen mode @ 2560x1440. Windows 8.1 x64
I can't perform a calibration. When I click on the Calibration tab all I get is a blank screen with ERROR in the resolution box. There are screenshots in my Pictures/Frontier Developments/Elite Dangerous folder.
Am I missing something here ? It Appears (according to the guide) that I click the Calibrate! button and it should ask me to browse for a screenshot to calibrate on.
As soon as I do this I get a fatal error message and the program closes.
Tried running as Admin, same result. Both the x86 & x64 Visual Studio Redistributables are installed.
Any ideas ?
The dump file is here http://goo.gl/NfSvQC (I RAR'd it as it was huge)
Cheers
Hi all, have been using EliteOCR and heard that this is a better solution so I thought I'd give it a try.......and I'm stuck on step 1
Running 1.83 with ED running in fullscreen mode @ 2560x1440. Windows 8.1 x64
I can't perform a calibration. When I click on the Calibration tab all I get is a blank screen with ERROR in the resolution box. There are screenshots in my Pictures/Frontier Developments/Elite Dangerous folder.
Am I missing something here ? It Appears (according to the guide) that I click the Calibrate! button and it should ask me to browse for a screenshot to calibrate on.
As soon as I do this I get a fatal error message and the program closes.
Tried running as Admin, same result. Both the x86 & x64 Visual Studio Redistributables are installed.
Any ideas ?
The dump file is here http://goo.gl/NfSvQC (I RAR'd it as it was huge)
Cheers
Sorry cant help you , DEV will, but post the Dump file link here in issues https://github.com/stringandstickytape/RegulatedNoise/releases with a brief of the problem , will help you and others out in the long run
Last edited:
Hi all. Sorry I haven't been around more... I'm trying to carve out time where I can :-(
Pretty-much. In fact it's impressive work by .NET that it doesn't crash - it's a divide by zero error. It's issue #16 at GitHub
Please retry with the latest version and post back. There were unfortunate bugs in the last couple of releases. Thank you for the dump files, these are enormously helpful, especially when pushed for time to replicate problems.
I did try this once and it didn't OCR well on a 2560x1440 screenshot :-/
There's a whole bunch of autocorrection and lookups and so on, but they can only go so far before we risk (say) replacing INDIUM with INDITE. Agreed that an additional manually-configurable autocorrect would help, i can't find this on the issues list at the moment though I did think it was there :-/
Spot-on, thx Lightning. You can reject the connection and it will work fine, but you won't be notified about new versions - which is pretty helpful when nasty little bugs have been creeping in...
Hoping these are side-effects of the bugs I've fixed in the last couple of days. Please retry the latest version and post back.
Can I just say thanks to everyone who's feeding back, suggesting ideas, providing bug reports and dump files, and most of all helping people out. These things mean I can at least use what time I have, as productively as possible. So thanks
Pretty-much. In fact it's impressive work by .NET that it doesn't crash - it's a divide by zero error. It's issue #16 at GitHub
henry1491 said:
JPB said:
H1ghland3r said:
deMangler said:
Please retry with the latest version and post back. There were unfortunate bugs in the last couple of releases. Thank you for the dump files, these are enormously helpful, especially when pushed for time to replicate problems.
I did try this once and it didn't OCR well on a 2560x1440 screenshot :-/
Neotribe said:
There's a whole bunch of autocorrection and lookups and so on, but they can only go so far before we risk (say) replacing INDIUM with INDITE. Agreed that an additional manually-configurable autocorrect would help, i can't find this on the issues list at the moment though I did think it was there :-/
Lightning said:
Spot-on, thx Lightning. You can reject the connection and it will work fine, but you won't be notified about new versions - which is pretty helpful when nasty little bugs have been creeping in...
malagon said:
Vapour said:
Hoping these are side-effects of the bugs I've fixed in the last couple of days. Please retry the latest version and post back.
Not clear what's going on here, if it can't read your resolution settings it should take them from the manual screenshot calibration. Thanks for the dump file, I'll try to have a look.pgriffith said:
Makes sense, I'll see what I can do (at least it's a quick one!).Snake Man said:
Can I just say thanks to everyone who's feeding back, suggesting ideas, providing bug reports and dump files, and most of all helping people out. These things mean I can at least use what time I have, as productively as possible. So thanks
I have another suggestion to improve OCR:Can I just say thanks to everyone who's feeding back, suggesting ideas, providing bug reports and dump files, and most of all helping people out. These things mean I can at least use what time I have, as productively as possible. So thanks
When I capture data the OCR is sometimes is having a hard time and screws identification up. When I manually correct the values and names, then it often seems to me that the OCR didn't manage to separate the digits/letters from each other. This is even more the case when I change the ingame FOV to larger numbers.
Now my suggestion:
Add an option in OCR-tab to add a value that "stretches" the screenshot horizontally before the OCR takes place. This should make it easier for it to separate the letters and digits and lead to less errors and false identifications. Maybe you can join this with the FOV setting/display (which is greyed out for me?): Enter the FOV number you have set in the game - and allow for greater numbers to further stretch.
- - - - - Additional Content Posted / Auto Merge - - - - -
I have placed it there for you, makes it easy to find it
Last edited:
What resolution are you running at? Care to post a sample screenshot?
I see what you're thinking, but you'd be surprised - any stretching of the image causes blurring (not necessarily visible to the naked eye) which makes the OCR worse, not better - I tried this out a little while ago.
You could try this for yourself - stretch out the image in MSPaint, recalibrate in RN, then copy and paste the screenshot into your screenshots folder so RN picks it up as a new screenshot. Let me know if you try this, but my reasonably-educated guess is that it'll make things worse, not better
I did read somewhere about turning off GUI effects somehow. You might like to investigate that - it's probably responsible for characters seeming to merge into each other, because it creates little drop-shadow effects and the such...