ATI Video Capture Frustration

I'm having difficulty with video capture using my ATI AIW 9700.

When I capture composite or S-Video inputs, I get this annoying horizontal line thing that looks like (and worse than) this:


I'm capturing/encoding the video to MPEG1 format. The lines appear in all the resolutions I've tried. Actually, I don't think it has to do with capture at all, because I see the same artifact when I'm just playing video. Can anyone help?

crap, attachment didn't work

That's related to interlacing, I think. Look in your options for things with the word "interlace" and mess with that.

www.dscaler.org

I tried using each of the three interlacing options that could be chosen in the MPEG capture setup. Each of them still produced the same result to greater or lesser extents. Your mentioning it does make me believe you are correct, though.

From the compatible cards list on the dscaler page:

ATI All-In-Wonder
The Tuner/Capture On This Combo Tuner/Video Card Will Not Work With DScaler

Darn. Looked like a good program.

I tried using each of the three interlacing options that could be chosen in the MPEG capture setup.

I don't know if the MPEG capture is a separate set of options from the regular video card settings on your card. But you said you also noticed the issue when simply viewing video images. So are there other options somewhere else in the video card settings, somewhere other than the MPEG capture settings?

Beyond that, I'd start looking for hardware/physical problems, like a bad video cable or a ground loop between your PC and your video source.

Argh. I forgot about that. Sorry...

I wonder if it's worth considering adding a low cost dscaler compatible capture card. (Assuming you have a slot free..)

The video display options are very limited (NTSC/PAL, connection). I'm not sure what to look for with respect to your other suggestions.

The lines are definitely the two fields that are interlaced togther to give you NTSC video. If you've researched, by now you know NTSC is composed of 2 fields, each updated at 30 frames per second. Gives you a refresh of 60Hz (some poeple think it's 60 FRAMES per second, but it's 60 fields and each field only contains 50% of the vertical information composing a frame). Blah blah blah... The vdeo you recorded includes as much information as could be extracted by the capture device (capturing at ~640x480). Removing the "lines" will in every instance involve some type of processing that will alter (I don't want to say degrade) the output. The most simple and ugliest would be to double each field. This gives of course a blocky effect. The OTHER, is to captuer at a lower resolution like 320x240...

When you are WATCHING TV through the ATI hardware it is generating the output through our hardware overlay. This is the "back-end" scaler of the chip. This back-end scaler (the video scaler) features all sorts of goodies for video processing *while displaying* Including adaptive De-Interlacing. So when you watch live, you don't see any lines.

When you play back the file you may do so in two ways... ATI's file player should use the very same overlay if it is not already being used by 1) the TV's live stream or 2) another video file or 3) a DVD. This will again de-interlace, real-time on playback. If you are already using the overlay then video will play using the front-end scaler (the 3D engine's scaler). You can think of the video as a texture being manipulated on a simple plane. The 3D engine doesn't feature hardware-based adaptive de-interlacing. So you see the lines. It's better than a pure software-based solution because it will scale smoother and faster in its basic form. It's less expensive to filter the texture using the ardware than it would be with a software-based algorithm.

Which brings me to my next point... It would be possible to implement a software-based de-interlacing algorithm to work in conjunctoin with the front-end scaler. This could possibly be done as an extension to Windows Media player or put into another player application (such as ATI's own). However, I think the best solution would be to write a pixel shader program to do the de-interlacing with the 3D hardware. This would be placed into the driver layer responsible for video acceleration.

Right now we are plagued with the same issue, but with DVD playback, in our Mac software. With the move to Jaguar (10.2) we've moved to using the front-end instead of our traditional overlay. This has many benefits for Quartz Extreme's interface, but now we lack the de-interlacing feature so many people loved for their broadcast-sourced DVDs. The video scaling abilities (tap filters, etc..) of the hardware overlay also exceed those of the 3D scaler (again, by design because of their intended purposes). Our Mac video accelerator component doesn't de-interlace either, but interlaced movie-file content is rare unless you're doing video captures (an we don't have an OS X capture solution right now)

I don't know what future drivers will be like for Windows, but I can tell you that I put in a feature request to implement the shader-based de-interlacing. Nice coincidence that you would have mailed me about the same issue.

Ok, I'm nursing a nasty head-cold, so I'm gonna go and pass out now.

Bruno

ATI's capture software already does de-interlacing of incoming video for playthrough and the video acceleration driver does it for playback. Only one instnec of the overlay supporting adaptive de-interlacing can be active at once though.

If you look on the DScaler site, you will notice they are working on a DirectShow filter. That would be the most appropriate way to use it with ATI's or anyone else's cards for playback. Of course anyone with the necessary video experience and with a DirectShow SDK should be able to make their own filter.

Bruno

Hey, Bruno...

I know that at first I said that it was related to the interlacing, but the more I look at that picture, the more I wonder if it's just deinterlacing artifacts. I mean, I know what a poorly deinterlaced image looks like, and that picture he posted looks significantly worse than that. It's almost like one of the two fields is damaged somehow. That's why a ground loop was listed in one of my suggestions... if there was 60-cycle ground loop noise being introduced on the input wire, couldn't it theoretically cause interference on only one of the two fields?

I mean, I can't imagine that all the fine ATI products cause an interlaced image to look that bad...

Thanks for the great response, Hugo!

I went back and looked at the settings again. This time I chose, under the interlacing options, to "Encode Interlaced." I know I had tried this option before, but now, for some reason, the resulting file did not have the problem I had before. Perhaps it was a combination of settings this time.

Regardless, your post was very helpful.

Ah. Well. Never mind, then.

You mean Bruno.

Damnit! You're right. Sorry about that

ps- I did know his name. just a momentary slip. he hasn't been around much lately

Not replying to anyone specifically here but does anyone have any opinons for gaming video cards? Specifically the 9700 versus the latest GeForce 4.

Every review site has opinions. And to sum up their reviews: 9700 hands down.

The funniest thing in a long time just happened though. Some industry analysts (I use "industry" and "analysts" VERY losely here) just awarded the GeForce FX a prize for the best graphics chip of 2002. They also awarded the Athlon 64 a best microprocessor award. Oh, one of their primary criteria for awards: Products must be commercially available in the year preceeding the award (that means available within 2002). Neither of those two products were available. Neither are even available now.

Bruno

Tony, that shot was a digital photograph of the monitor, so I attributed some of its characteristics to that fact. If it was taken during a quick pan or other large change in the overall frame, you'd be able to see the lines in a very dramatic way. Also, if the image is frozen like that, the lines are much more pronounced than when you're actually watching. Look at the Titanic VHS capture on the DScaler site, it looks pretty bad as well.

So it turns out there was some encoding problem and it wasn't just a matter of de-interlacing on playback. Interesting. I haven't done any captures to anything but the default VCR format yet. Mainly because my captures were for my own "time-shifted" viewing purposes. When I add a DVD burner to my network then I'll be more concerned with the other options.

So I was scrolling the page up and down to see where I missed Hugo's post... Ha. I didn't know my reply was so authoritative that I'd be confused with him.

If any other oddities come up or if anyone had any neat feature/option suggestions let me know. I'll submit them internally (and take credit for them ) I'm also working on my own design document for this type of application.

Bruno

Ha, sorry about that again.

Yes, you're right. I that picture is about as bad as it was getting. I purposely chose an instance where it would be most pronounced, so I could clearly show what I was talking about. That shot* was a fast zoom in, pan up, with strobe and lighting effects. Doesn't get much tougher than that

The main suggestion I would give, Bruno, would be to give up all this "skin-like" crap, and give me normal freakin' windows! Why does the window have to have its own minimize/maximize/close buttons and fancy border? It just looks cheesy and cheap, because the edges are pretty jagged.

My other suggestion would be to offer an option to store the video and sound files seperately when encoding. Currently, I'm using Impression to create my DVDs, and it requires that the elementary streams be imported seperately.


*ps-the sample shot is from a video I recorded of a great moment on Beat the Geeks where the TV Geek tells a contestant that he had just slept with the guy's dead mother. priceless. God I miss that show.

That shot* was a fast zoom in, pan up, with strobe and lighting effects. Doesn't get much tougher than that

AH, that's cheating! You didn't tell us that!

I thought it was a static shot of a set of game-show contestants, so I couldn't understand why one field was significantly darker than the other field. I didn't realize there was movement and strobe lighting. That explains the fact that I thought one field looked bad enough to be "damaged".

Hehe, quite deceptive, wasn't it?

You really could see it on static shots as well. I had a clip from Conan Obrien, and part of the set over his desk are these large, spherical lights. Towards the top and bottoms, this effect was VERY noticeable, moreso than the average effect of pixelated curved edges.

But basically, it was confined to any movement in the vertical direction.

That really is a terrible shot. I'd be embarressed to encode a shot with a guy with such a mullet as the guy on the right.

Oh man, it's a shame you haven't seen this show. Of course, it would mean little to you, not being from the US, as the whole show is primarily about US pop trivia.

Alas, I don't get to see it anymore either. It was cancelled recently

it would mean little to you, not being from the US, as the whole show is primarily about US pop trivia.

Several times now at USENIX conferences the quiz show done as the last event has had non-U.S. contestants in games dominated by U.S.-centric trivia who have done quite well.

You never know where someone's interests lie.