ATI AIW 7500 AGP Radeon Capture Big Problem

[theseeker2]

I am still curious if anyone has any feedback on the circle test with the drum above

[lollo]

From what I've read, the Diamond VC500 USB captures the full range of 0-255

Different report from several users, maybe because production/drivers change, but it should capture 14-245 range. And in term of picture quality it will not be better than a AIW or a 600 USB.

It may suffer of AGC variations, even this differently reported from several users

I am still curious if anyone has any feedback on the circle test with the drum above

What are yours? Did you measure with a ruler its geometry (I measured a less circolar shape for AIW, but it was a rush)

[dellsam34]

OP, If you have an analog camcorder shoot a perfect circle perpondicularly and capture the footage with both cards and post raw samples here for comparaison, or at least unprocessed raw screenshots from the video stream, not from the screen.

[Sharc]

I am still curious if anyone has any feedback on the circle test with the drum above

I gave my comment in post #23 on that "circle" test. Pretty much a vain effort. You would have to find a picture with a true circle taken right from the front rather than slightly from the side. It may not exist on that video though. So why care if one is undecided?

[Sharc]

The histogram works, but there are spikes and gaps whenever I make even a tiny adjustment.

That's normal. The 0....255 steps are integer numbers (quantized). There are no fractional steps.
Spikes indicate accumulated code values on the 0....255 integer scale
Gaps indicate missing code values on the 0....255 integer scale.
When you apply a gain to a signal and do the integer maths you will find that some steps may become unused (gaps) and other steps get used more than once (spikes).
(I think lollo supplied a demo farther up.)
In the extreme case it can lead to visible banding or posterization. Unlikely for noisy VHS video though.

[theseeker2]

The histogram works, but there are spikes and gaps whenever I make even a tiny adjustment.

That's normal. The 0....255 steps are integer numbers (quantized). There are no fractional steps.
Spikes indicate accumulated code values on the 0....255 integer scale
Gaps indicate missing code values on the 0....255 integer scale.
When you apply a gain to a signal and do the integer maths you will find that some steps may become unused (gaps) and other steps get used more than once (spikes).
(I think lollo supplied a demo farther up.)
In the extreme case it can lead to visible banding. Unlikely for noisy VHS video though.

Does this seem normal with the ATI 600 USB? All I did was adjust the sharpness down to 0 as recommended and here's what the histogram does.


[Attachment 71143 - Click to enlarge]

[Sharc]

Does this seem normal with the ATI 600 USB? All I did was adjust the sharpness down to 0 as recommended and here's what the histogram does.

I don't know. Squeezing or expanding the levels can introduce periodic spikes and dips in the histogram. But changing the sharpness?

[lordsmurf]

, thare is no need to shink 235-252
Better to handle the levels adjustement, if needed, in post-processing or at encoding stage.
However, capturing 16...235 is safe for sure!

Most hardware designed for consumer analog transfer adheres to legal 16-235 levels, as it should. So you either correct levels in proc amp (and proc amps are not "rudimental"), or you allowing values to be clipped. Clipping is not bad, but rather input is invalid/illegal/bad. And not correcting can be simply lazy and uncaring.

I am still curious if anyone has any feedback on the circle test with the drum above

I gave my comment in post #23 on that "circle" test. Pretty much a vain effort. You would have to find a picture with a true circle taken right from the front rather than slightly from the side. It may not exist on that video though. So why care if one is undecided?

Yes, but no. A circle is still circular until you start to get too obtuse in the angle. Angular difference isn't the same as aspect augmentation, though it will be as you get too far from true. I just don't see that in the sample image posts. I think a circular comparison is warranted here. Use Photoshop to generate a circle on the post-capture footage screen grabs. For pre-capture, do as suggested above, use camera to shoot circle, but noting it will add lens distortion; test pattern better here.

Timmy like apples.
Timmy like blue.
Apples must be blue?

^ Don't assume. Verify. Test, experiment to arrive at conclusions.

[lollo]

Most hardware designed for consumer analog transfer adheres to legal 16-235 levels, as it should

Not really, Bogilein docet with facts: https://www.digitalfaq.com/forum/video-capture/9865-canopus-advc-110-a.html#post62640

and proc amps are not "rudimental

Inside the capture card, they are, compared to post-processing (with dithering, eventually), because they act on a signal already digitized.

Their level manipulation introduces spikes and gaps, as shown and experimented in this thread https://forum.videohelp.com/threads/402267-Adjusting-capture-card-levels-not-sure-if-it-s-right; with AviSynth you have a better results (*)

So you either correct levels in proc amp (and proc amps are not "rudimental"), or you allowing values to be clipped.

That's what we've been saying forever. The only difference between Sharc and me, is that he prefers to adjust the levels to 16-235, I prefer to adjust the levels to the range of the card, in order to avoid unnecessary level correction performed by the procamp (because what explained in (*))

If something is not clear, just ask.

[Sharc]

What do you guys see on your PC and TV?

[lordsmurf]

Most hardware designed for consumer analog transfer adheres to legal 16-235 levels, as it should

Not really, Bogilein docet with facts: https://www.digitalfaq.com/forum/video-capture/9865-canopus-advc-110-a.html#post62640

Something in that test is bogus, as sanlyn discusses. ATI AIW doesn't clip values -- although I'd not rule out "x" gen drivers causing issues, as they already do in other areas.

and proc amps are not "rudimental

Inside the capture card, they are,

Ah. Yes, capture card "proc amps" are as you state. Not actual proc amp, and post-capture is generally no different than those card settings.

Their level manipulation introduces spikes and gaps, as shown and experimented in this thread https://forum.videohelp.com/threads/402267-Adjusting-capture-card-levels-not-sure-if-it-s-right; with AviSynth you have a better results (*)

Yep.

So you either correct levels in proc amp (and proc amps are not "rudimental"), or you allowing values to be clipped.

That's what we've been saying forever. The only difference between Sharc and me, is that he prefers to adjust the levels to 16-235, I prefer to adjust the levels to the range of the card, in order to avoid unnecessary level correction performed by the procamp (because what explained in (*))
If something is not clear, just ask.

Clear now.

[lollo]

Something in that test is bogus, as sanlyn discusses.

Sanlyn was wrong there, because confuses MPEG test pattern as source for analog capture beeing not appropriate, while it was just a "test pattern", and he always misunderstood the "HDMI capture approach" technique from the German guys.

The results from Bogilein speak for themselves; I was able to reproduce some of them, but I do not have all the capture cards he own, so I trust him for what I cannot reproduce.

Clear now.

I apologize if sometimes my english is not good enough

[Sharc]

FWIW here another demo showing how extreme level manipulations affect the shape of the histogram (top vs bottom picture).

[lollo]

Thanks Sharc. This question about input levels, 16-2XX range, procamp settings, etc. is asked very often, so more examples we have, better we can explain.

[Sharc]

I am still curious if anyone has any feedback on the circle test with the drum above

I gave my comment in post #23 on that "circle" test. Pretty much a vain effort. You would have to find a picture with a true circle taken right from the front rather than slightly from the side. It may not exist on that video though. So why care if one is undecided?

Yes, but no. A circle is still circular until you start to get too obtuse in the angle. ...... Don't assume. Verify. Test, experiment to arrive at conclusions.

I would have to disagree. What is your estimation of the camera angle to the drum? 12 degrees tilt sideways give a circle distortion of ~2.2%. Same as the 704 vs 720 discussion, hence I am sceptical that the circle test proves anything here, especially when considering that the deviation from 704 or 720 is ~+/-1% only if the "true" width is 711 pixels for that chip as stated by Brad.

[theseeker2]

Speaking of testing, I tested out the ATI USB 600 on another machine, and it did the exact thing thing with the histogram when I turned the sharpness down to 0.

If anyone else who owns a ATI USB 600 could run a simple VirtualDub histogram test just to make sure this is normal behaviour when turning down sharpness to 0, I'd greatly appreciate it! It's not too hard for me to acquire another one if this one is defective.

[oln]

The camcorder (or another VCR for that matter) may not even record a circle entirely correctly in the first place so it's kinda hard to base anything on that IMO. If you want to check to what degree it's stretched etc it might be better to test it on some test card dvd output that shows the exact pixels.

[theseeker2]

I just found a recent thread comparing the picture quality of an AIW 9600 to a USB 600 and comes to a similar conclusion that you all helped me discover: that the USB 600 provides a sharper more detailed image and also that AIW 9600 slightly stretches the image: https://www.digitalfaq.com/forum/video-capture/12424-usb600-better-quality.html. This makes sense to me since the AIW 9600 also uses a Rage Theater chip just like my Radeon 7500.

I'm still hoping to figure out the mystery of the sharpness setting on my USB 600 with the gaps in the histogram. I just compared two captures taken from a professional recorded VHS tape, one with sharpness at 0, and the other with sharpness at the default of 2, and I'd be curious to know what you all thought.

Here's the image slider: https://imgsli.com/MTgwNDk0 (thanks lollo for teaching me about this!)

USB 600 at sharpness 0:

[Attachment 71192 - Click to enlarge]

USB 600 at sharpness 2:

[Attachment 71193 - Click to enlarge]

[Sharc]

Obviously the sharpness setting does not only affect the sharpening but also changes the levels (brightness, contrast) slightly. Therefore the differences in the histograms. This is just how this card and driver seem to work. Nothing wrong with the device I think.
The final result will depend on your post processing anyway.
(VirtualDub makes a YUV->RGB conversion, so you can't directly compare the histograms)

[Skiller]

Sharpness at 2 has coarse "dither-like" posterization all over. Wouldn't use 2, no matter what the histogram does when setting sharpness to 0.
Sharpness at 0 looks good and natural, much better.

[lollo]

Here's the image slider: https://imgsli.com/MTgwNDk0 (thanks lollo for teaching me about this!)

It does not replace a check on a full video with motion, img slider is useful just to check some aspect of the video while comparing.

Concerning sharpness setting impact on the histogram, is normal, because behind it there is a level manipulation anyhow, as explained by Sharc and Skiller.

About the quality of the capture, given the condition of your tapes, and the good capture cards you alread have, the only improvement can be in the VCR (player), which is the most important element in the chain. If you already have a high end S-VHS VCR you're done.

Enjoy your captures!

[Sharc]

Yes. And both show missing codes in the waveform, so there is not much difference in this respect. The sharpness=2 however crushes/clips the blacks slightly at Y=16 (maybe it's from the head switching crud only, not that this really matters).

[Skiller]

Back in the day with my SAA7134 based PCI capture card, the default Proc-Amp settings were not neutral either and produced a compromised histogram. I then played with the Proc-Amp settings until I found the neutral combination of settings that gave a smooth histogram. I found that Contrast expands and compresses the histogram whereas Brightness just shifts it as a whole, as to be expected (but your mileage may vary).

No surprise this did not result in perfectly correct levels, so I connected a DVD-player playing a suitable test pattern to determine a reference point of what correction, plus-minus, in AviSynth (SmoothLevels!) is needed. Also no surprise, nothing was clipped or missing that would not have been clipped anyways with different Proc-Amp settings.

In the end this gave me the same levels but with a smooth histogram and smooth gradients. Certainly not ultimately needed with noisy sources, but you only need to do it once.

[theseeker2]

You all have helped me tremendously, I can't thank you enough! I am moving on with ATI USB 600, sharpness set to 0, and my S-VHS VCR for capture. Also looking forward to capturing some Hi-8, Beta, and a few other things I've acquired over the years

Best to all of you!

[lordsmurf]

You all have helped me tremendously, I can't thank you enough! I am moving on with ATI USB 600, sharpness set to 0, and my S-VHS VCR for capture. Also looking forward to capturing some Hi-8, Beta, and a few other things I've acquired over the years
Best to all of you!

ATI 600 USB is a fine card.

ATI AIW is worth the effort, but there are driver, card revision, and card model issues to overcome. It's generally not as plug-and-play as the 600.

Remember to run the audio hack for the 600, pull audio down to a 50% value (128). The default 100% easily red-lines and distorts.

[Sharc]

Back in the day with my SAA7134 based PCI capture card, the default Proc-Amp settings were not neutral either and produced a compromised histogram. I then played with the Proc-Amp settings until I found the neutral combination of settings that gave a smooth histogram. I found that Contrast expands and compresses the histogram whereas Brightness just shifts it as a whole, as to be expected (but your mileage may vary).

No surprise this did not result in perfectly correct levels, so I connected a DVD-player playing a suitable test pattern to determine a reference point of what correction, plus-minus, in AviSynth (SmoothLevels!) is needed. Also no surprise, nothing was clipped or missing that would not have been clipped anyways with different Proc-Amp settings.

In the end this gave me the same levels but with a smooth histogram and smooth gradients. Certainly not ultimately needed with noisy sources, but you only need to do it once.

Interesting approach.
Just a footnote: Our efforts for obtaining a smooth YUV histogram may be flushed by the YUV (limited range) to RGB (full range) conversion of the graphics card driving the PC or TV monitor (RGB devices). Expanding the 16....235 luma to 0....255 introduces a gap every 12th or 13th step which is visible in the attached (grayscaled) picture as regular sharp gaps (zero value) in the RGB histogram. The good point is that these gaps do not mean a loss of data, opposite to peaks we would get when squashing the histogram.

[Skiller]

MadVR (DirectShow video renderer) does a high quality conversion from limited range YUV to full range RGB in high bitdepth and with dithering. Highly recommended.