AviSubDetector v0.6.0.5

[bshater]

can someone show me how to use OCR?

[ericf]

Any development in the OCR?
I keep getting hang-ups with it and it stops reading and writing the characters. Even if I restart the program it's no good. How much memory does it need? I have 1000 MB but mabye that isn't enough? Or is it not memory related? I can see that the OCR function starts to write the letters but it stops before the sentence is finished. That's why it doesn't write any letters at all. Manually typing letters IS an option but not one I really feel is supposed to be necessary. Of course, letter types like Times New Roman are more difficult for the program to read than Arial but does that mean it won't agree at all with those types of letters?

The features of the program Subrip are truly not acceptible so that is not an option for me. They simply don't work at all! However, I'm having a lot of problems with AviSubDetector as well and I'd like to know if something is about to be changed to the better.

Thanks for making the program.
ericf

[Shalcker]

Any development in the OCR?
I keep getting hang-ups with it and it stops reading and writing the characters. Even if I restart the program it's no good. How much memory does it need? I have 1000 MB but mabye that isn't enough? Or is it not memory related?
It IS memory-related. But it's not about memory amount - it's about memory fragmentation (and Delphi memory manager). The way OCR works now leaves a lot of small memory fragments, which aren't reclaimed by default memory manager, resulting in "out of memory" error even though total amount of "free" memory is relatively large... and once that "out of memory" happens OCR stops (usually mid-sentence due to masked access violation while it tries to add unrecognized letters), and program can ocassionaly freeze too :)

Are you using latest version BTW? This error shouldn't pop there too often...

I can see that the OCR function starts to write the letters but it stops before the sentence is finished. That's why it doesn't write any letters at all. Manually typing letters IS an option but not one I really feel is supposed to be necessary. Of course, letter types like Times New Roman are more difficult for the program to read than Arial but does that mean it won't agree at all with those types of letters?
As long as letters aren't too disjointed, letter type shouldn't influence anything. Most problems can be resolved by proper settings for color masks and limiters - which is easier to explain on example. If you have any image which gives you trouble with OCR, you can post it and i'll explain which settings are optimal for such subtitles (and how can you deside what should be changed from defaults).

[ericf]

"Shalcker:
It IS memory-related. But it's not about memory amount - it's about memory fragmentation (and Delphi memory manager)."

Oh... I don't understand but okay.

"Are you using latest version BTW? This error shouldn't pop there too often..."

I think so. AVISubDetector0.6.0.5. Oops, there's a 0.7. version out. I'll try that.

"As long as letters aren't too disjointed, letter type shouldn't influence anything. Most problems can be resolved by proper settings for color masks and limiters - which is easier to explain on example. If you have any image which gives you trouble with OCR, you can post it and i'll explain which settings are optimal for such subtitles (and how can you deside what should be changed from defaults)."

I have had problems with most types of subtitles. Unless they are white with large black borders. One problem is that while many subbers use the same letter types they keep changing the outline color. How do I make a setting that can deal with that?

Even if I use avisynth to make the files black and white there seems to be sufficient differences to cause problems.

I'll try to include an image with several problematic subs below. I don't know much about adding images to boards.

This avi is a dvix3 low motion clip. I can't seem to use it with this avisynth script:

AviSource("c:\AmoviesDVD03\MegamiKouhoseiCFGep03.avi")
Lanczos4Resize(704,480)
GreyScale()

Can you tell me why (I only get a black image) and do you have some advice on how to be able to make all these OCR well with one setting and even when the background is bright or full of little details? The green subs only work when the background is rather dark.

ericf

PS: The new version seems to do a better job of ORCing this file. However, I have to OK every line before it moves on. Is that how it's meant to be?
Thanks.
ericf

[maxleung]

Some serious problems to report:

The frame offset reported for each detected subtitle are incorrect. For example, if a subtitle is reported on frame 9333 (as shown in the current frame textbox), the script will incorrectly write the line as starting on frame 9318.

The result is that the subtitles go completely out of sync with the video!

The video is reported as 23.976 FPS. The video is in FourCC XVID format. The AVS script looks like this:


directshowsource("[Ani-Jiyuu]_Ergo_Proxy_-_10_[97428AD2].avi",audio=false)
ConvertTORGB24()


Is there a setting I missed?

I tried using Subtitle Workshop to change the FPS rate but that didn't work - still a lot of places where the subtitles are off by 2 seconds or more!

[maxleung]

Another bug:

If you save the script in SSA format, and there are line breaks in the text, this will result in a bad SSA script. Line breaks should be replaced by \n or \N automatically.

Also, if you read back an SSA script written by AviSubDetector, and if any subs span more than one line, the extra lines will be lost! This lost me 2 hours of work the other day - I was loading an SSA script created by ASD earlier, then saved it again, losing all the subs that spanned more than one line.

Workaround is to never add linebreaks to a subtitle (use \n instead), or perhaps save in SRT format <- but I haven't tested this.

[maxleung]

The weird thing about the incorrect frame problem is that the durations seem to be correct most of the time. From my attempts to correct the sync using Subtitle Workshop, it seems that ASD occasionally loses 5 or more frames every now and then. The effect appears to be cumulative in most cases.

[maxleung]

I think I figured out why the subtitle detection goes out of sync - whenever a scene goes too dark or too bright, it loses frames. I'm guessing that it loses synch depending on how long the scene stays dark or bright. This could be a codec-related problem - the frame counter doesn't include these dark/bright frames!

[maxleung]

Heh, seems like I'm the only one here. :)

I fixed my problem - using the DirectShowSource function was the problem. When I use AVISource and ensure that FFDShow's XVID decoder was enabled in its VFW configuration fixed it! No more frame synch problems! :D

(BTW, ASD is a really great tool!)

[ericf]

I'm getting quite frustrated here.
I've been trying to come up with a good setting for an anime file with white subs and black borders. Okay. The settings seem pretty perfect but the OCR just keeps marking everything around the letters. There's nothing to OCR outside the text.
What gives? Why does this happen?

And why is it impossible for me to use a setting I used one time in OCR color (not using auto color other than analyzing it first) the next time I open the program? Using those settings the text isn't read correctly. It disappears into oblivion. Either the OCR window is gray, black or pink.

Can anyone help me with this?

And what do I do when all the words stick together in every sentence? Theysticktogetherlikethis:
andsometimes.there's.alotofdots.inthetexteventhoughtherearenone.inthetexttheprogramissupposed.toocr.

Has anyone worked out how to explain the functions for a layman?
I simply cannot get the terms used here and in the program. Both because the language isn't my first language and because it's technical language that seems to define things I couldn't understand anyway.

This thread seems quite dead but I hope someone reads it.
Thank you.
ericf

[Shalcker]

I'm getting quite frustrated here.
I've been trying to come up with a good setting for an anime file with white subs and black borders. Okay. The settings seem pretty perfect but the OCR just keeps marking everything around the letters. There's nothing to OCR outside the text.
What gives? Why does this happen?Can you put an image somewhere for checking?
Setting colors directly to match borders and subtitles should help (if you aren't using auto colors). Check that their types are right - text is marked as Tb on OCR/Main/Colors tab and outline is marked as Oc. Click on them there to change their type.

And why is it impossible for me to use a setting I used one time in OCR color (not using auto color other than analyzing it first) the next time I open the program? Using those settings the text isn't read correctly. It disappears into oblivion. Either the OCR window is gray, black or pink.
Well, these settings aren't saved, so they aren't restored either... adding that save is relatively easy though, do you need it? :) Can be done in a few hours...

And what do I do when all the words stick together in every sentence? Theysticktogetherlikethis:
andsometimes.there's.alotofdots.inthetexteventhoughtherearenone.inthetexttheprogramissupposed.toocr. Well, pretty much any small area looks like a dot. So it often gets a good match rating, inserting dots from "noise" above or below letters.

As for missing spaces, that is controlled OCR/Main/OCR Settings/Space Distance. Just reduce it and they should work more or less fine (with exception to most italic fonts).

[ericf]

Thanks for answering.
I'll post two images here. It tends to take a while until they appear but they will be here (like the former image I posted with several different outline colors in one episode).

And if I have a particular color (perhaps green) as outline, do I take a still and go into Photoshop and find out the RGB of that color and set it in the OCR Colors window? 255 and 100 makes a white background. 0 and 30 makes a black background. What does the 100 and 30 mean?

Using outline as black and text as white doesn't make the OCR read any text at all in the below pictures. Why? 3rd image at bottom. Do I have to find out the exact color combination in Photoshop or some other program to make it work?

The same question goes for Color Domination / Use Color Processing in the settings window. If I decide to use Use Color Processing, what PreProcessing Colors boxes should be ticked? All? None? What does X mean? T should be Text and O outline. And I don't know what Shift+LMB means when you mention it at Separation by subtitle color. What does it do?

I usually use Color Difference / Inverted Blocks / Inverted lines in the setting window.

I'm sorry if I seem stupid. I've tried to make some changes to settings in order to make the OCR work properly but I don't seem to be making it very well.

I've fiddled around with the settings in Auto-Color sometimes as well.The original 6-100-64 settings. I've found that it's sometimes good to make the settings so that a blue outline appears around the characters. But sometimes that doesn't work.

Any thoughts on how I can improve that?

And making the settings you used last time stay when you open it the next time would be very nice.

And I understand the trouble with italic fonts. You can never make a j or i work properly since they can't be read separately.

Why does the box that I thought was there in the OCR window to tell me what the program is checking in the text so often end up outside the text area? I still get letters in the three boxes but isn't the black outline supposed to show the area it's OCRing?

One more thing.
I often get OCR sentences with the words of the last part of line 1 and the first part of line two something like this:

I never knew what hit me hu
e ntil told me I had been struck by lightning.

Can I avoid this in some way?

I always get Access violation at adress ************ when I try to exit the program. I have XP Home on my computer. Any idea what might cause this? (Pressing stop and removing the avi makes the program happy. But could you perhaps add a buton for exiting the program?)

Also, is there some way I can make the OCR not mistake ´ for ' or ,? It seems I get a lot of commas instead of apostrophes and accents instead of commas.

Sometimes I can't save the subtitles and symbols. The buttons work but I don't get a window for saving. It's when the whole file has been OCRd. The subs don't appear in the temp folder either. Is there any way to make sure the subs always get saved? Perhaps continually during OCRing? When I start the program and try another OCR the program always asks for me to choose wether to continue on the last subtitles or not. That way I can save the subs. I can't ever save the symbols I wrote into the symbol file, though. Is it possible to save that continually as well?

Thanks
Ericf

[renoturks]

Sorry to bother , but i need help, i'm having the problem:

invalid floating pont operation

what should i do

[Adub]

I recommend using Subrip. It is more updated and easier to handle.

[ericf]

Sure, it's more updated. Last time in 2006.

Anyway, the way it reads subtitles is in most ways a faulty process. AviSubDetector always finds the subs (except for some of the very short time subs that you have to make special settings for) with almost no change to default settings.

In Subrip, no two letters look alike. That means that every single one of the letters of a movie has to be entered manually. Not so with AviSubDetector. There may be a few duplicates of letters in the symbol sets but it reads characters very well.

No manual is a problem. I still don't know why things work when they do. But it is much better than Subrip.

No contest.

ericf

[help]

I can't even preview the .avi file:

Video: XVID 696x400 25.00fps 1148Kbps
Audio: MPEG Audio Layer 3 48000Hz stereo 128Kbps

[ericf]

Hello.
I have a problem with the symbols I saved.
Now the symbols are treated as garbage. Instead of the correct letters I get all kinds of strange symbols.
Can anyone tel me how to fix that?
I don't like to create a new symbol file every time I start OCRing subtitles.
Thank you.

[ericf]

I need to know how to open mkv/mp4 or any h264 encoded video in Avisubdetector.

I'm unable to open even a normal avi using this script:
AviSource("file.avi")

Thanks.
ericf

[Adub]

I'd try and help, but I can't seem to find a working link.

Edit: Ah, wait, videohelp has one. Give me a couple minutes.

Edit2: Huh, ok. I can load a simple avi source just fine with an Avisynth script.

Example script:

Avisource("myavi.avi")
ConverttoRGB24()

Just load that and you should be fine. It seems the program only accepts RGB24 colorspace, so you have to add that last line to all of your scripts at the end, but it seems to work fine.

[ericf]

It seems like I didn't have Avisynth (the common one) installed.


I thought that came with my codec pack.
Sorry for the trouble.
ericf

[Adub]

I would highly recommend against using codec packs.

FFDshow is all you really need. However, if you absolutely must use a codec pack, stick with CCCP.

[simonrule]

can any one help me how i create new symbol

[videoslicker]

Great program.

I've gotten great results when adding a brightness/contract filter to the AVS/VDR script. I'm ripping the subs off a jdrama fansub with clear outlined white subs. I turned the brightness all the way down, and the contrast a little up. It really takes away the video and accentuates the hard subs. I'm going to add a gamma filter to see if it helps any further.

DirectShowSource("D:\video.avi")
Tweak(bright=-255,cont=4)
ConvertToRGB24()



-- edit --

Actually this was terrible advice...

AVISubDetector needs to be able to recognize the colors in order to detect the subs. Just leaving the color/contract alone worked great.

There were bugs in the program that made it almost unusable but there were ways around it once you get familiar w/ the program. One major bug was that the program always quit going to auto mode after recognizing italicized letters. To work around this, pause during any italicized sub; DO NOT OCR that line; go to manual mode and manually enter in the subs until all the italicized lines are transcribed and then revert back to OCR (experimental).

Some helpful tips that helped w/ my sub rips...
- dont load symbols; start from scratch on new projects
- manual mode any scene that contain several artifacts that detect as subs
- manual any multi-colored outlined subs; the program only picks up 1 outline color at a time
- always have the preview window open so you can see the entire scene

[quz]

how to open vob file in asd?

[videoslicker]

how to open vob file in asd?

create a AVS script to frameserve the vob.

Open VOB files in Avisynth
http://forum.doom9.org/archive/index.php/t-109454.html

[Gfy]

If someone is looking for it, the source code can be downloaded here:
http://web.archive.org/web/20071031111130/http://animeburg.omake.ru/avisubdetector.htm

[TheFrenchDumb]

i love this tool, but i got some crashes, I dont use OCR.


Most important thing is the video size, if you got a too big image size (like 1080) it will crash.

I always convert in 700x400 (dont remember exactly) w


If you got new gen file (mkv/mP4)i suggest to convert into avi xvid+mp3

[djmasturbeat]

I can't open any avi files without error:
"Error in FrameOpen - conversion to RGB24 not available. Check that you have required codec installed."

Can anyone tell me what I need to do to get this working?
I have ffdshow installed, is there a setting I need to change there?

thanks

oh, I am on Windows 7 x64, also
maybe this won't run on it properly?

[Abyssal]

I can't open any avi files without error:
"Error in FrameOpen - conversion to RGB24 not available. Check that you have required codec installed."

Can anyone tell me what I need to do to get this working?
I have ffdshow installed, is there a setting I need to change there?

thanks

oh, I am on Windows 7 x64, also
maybe this won't run on it properly?

Hey,

All you need to do is to create an AviSynth script that calls the video file and converts it to RGB24.

I'm gonna assume you know that the program AviSynth exists to begin with so let's move on.

Just create a new text document and add the following:


AVISource("xxxx.avi")
ConvertToRGB24()


You obviously have to change xxxx.avi to whatever your file is called. To make it easier save this document in the same folder as the video file so you don't have the longest file path ever.

So anyway, you save this document with just those two rows as xxx.avs - So if you wanna call it starwars then save it as starwars.avs

Then you just open up this .avs file instead of the video file in AviSubDetector.