search

ali1234 / vhs-teletext

Software to recover teletext data from VHS recordings.
GNU General Public License v3.0
179 stars 21 forks source link

No teletext data from deconvolve #7

Open andrewferguson opened 7 years ago

andrewferguson commented 7 years ago

For some reason I am not able to get any teletext data from the VBI files. I've captured the data through the vbicat /dev/vbi0 > capture.vbi then tried to run deconvolve -H -S 20 -T 4 capture.vbi > headers.txt. However, once that completes, it states that: 0:01:02 : 12001 lines, 193/s total, 0/s teletext, 100% rejected.

I assume this to mean that it wasn't able to find any teletext data in the VBI data I captured. This is probably due to an issue with my setup, but I can watch the video contents of the VHS tape fine using xawtv, so it is not a driver issue. Additionally, vbicat accepts the data from /dev/vbi0 fine, and only occasionally displays a Frame dropped? message.

I've tried several different VHS tapes, but no luck with any of them. I've uploaded a sample VBI file which is from a BBC 2 program that is definitely before 2012, in the hope that some kind person who has a known working setup can confirm if it is an issue with my hardware, or my software setup. (My video player is a cheap Matsui from the late 1990s, so perhaps I need a better player?). When I try and run deconvolve on this particular VBI file, I get the following output:

No module named pycuda.driver
CUDA init failed. Using slow CPU method instead.
No module named pycuda.driver
CUDA init failed. Using slow CPU method instead.
No module named pycuda.driver
CUDA init failed. Using slow CPU method instead.
No module named pycuda.driver
CUDA init failed. Using slow CPU method instead.
0:01:02 : 12001 lines, 193/s total, 0/s teletext, 100% rejected.

I have attached the headers.txt file that this generates, in case it is of use to someone more knowledgeable than me.

I am really excited to discover this project, and hope to begin retrieving teletext pages soon. Huge thanks to the developers of such a fantastic project.

ali1234 commented 7 years ago

Turn off image processing on your VCR.

andrewferguson commented 7 years ago

Unfortunately I am not able to turn off image processing (the VCR has no remote, and even if it did I am not sure that would even be a setting). I tried another VCR today (Philips DVD / VCR combi) but that didn't work either. I have the remote for that, so I tried to find an "image processing" setting but couldn't.

I will now look for another VCR to use... what sort of model do you recommend (early 1990s?, late 1980s?). High end or cheap? Out of curiosity, what model are you using?

And just to check, I should be connecting the composite lead to the capture card, not the RF lead?

Thanks for all your help.

On 26 April 2017 at 12:30, Alistair Buxton notifications@github.com wrote:

Turn off image processing on your VCR.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/ali1234/vhs-teletext/issues/7#issuecomment-297368527, or mute the thread https://github.com/notifications/unsubscribe-auth/AM5MrTtpw12zdf-ksybu1tuHyCifuTnzks5rzyq6gaJpZM4NIFxI .

ali1234 commented 7 years ago

Use vbiview to display the raw capture data:

vbi-good vbi-bad

First is a normal capture, second is your capture. It looks like something ramped the contrast up to maximum and that is why it doesn't work.

ali1234 commented 7 years ago

Try this:

./deconvolve -r 24 -- capture.vbi

You can see it kind of works, but the signal is too damaged to make much sense out of.

andrewferguson commented 7 years ago

I've tired a few more VHS players, as well as a few different tapes, and have got it working now. (One thing that never occurred to me until a few days ago is that some VCRs are not very good at capturing the VBI data. Trying an older VCR as well as a tape that was recorded in an older VCR made the difference, and I was able to get a pretty good capture).

The only thing is I am getting quite a high rejection rate (97% rejected) when scanning the headers of the VBI data. This drops to around 25% rejected when deconvolving the full data. Are these sorts of rejection rates normal?

On 26 April 2017 at 20:52, Alistair Buxton notifications@github.com wrote:

Try this:

./deconvolve -r 24 -- capture.vbi

You can see it kind of works, but the signal is too damaged to make much sense out of.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/ali1234/vhs-teletext/issues/7#issuecomment-297522209, or mute the thread https://github.com/notifications/unsubscribe-auth/AM5MrRL_hiY2XiPoVgWMZNUYTtLxaRRrks5rz6CGgaJpZM4NIFxI .

ali1234 commented 7 years ago

High rejection rate is normal when header scanning because it skips 19 out of every 20 lines in order to get a high level overview of what is in the data. So you will see at least 95% rejection because of those skipped lines.

The purpose of this is if you have a tape with lots of different recordings on it you can quickly see what line number each recording starts/stops on, so you can split them out using the --start/--stop/--count options when doing a full pass.