AVS › AVS Forum › Video Components › Home Theater Computers › FFdshow lancosz resize...no difference!?
New Posts  All Forums:Forum Nav:

FFdshow lancosz resize...no difference!?

post #1 of 60
11/14/06 at 8:27pm
Thread Starter 
Hey,

Up to a few months ago I was really into getting the highest quality picture possible. I was using ffdshow with the lacosz resize option on parameter 2, resizing to native resolution of the screen(s) I use.

One day I decided to check and see if I could tell a difference. I took a screenshot of lancosz resize in full screen, and then a screenshot of media player classic doing the resizing by itself (no resize in ffdshow). I put them both in photoshop, at 100% zoom, placed the images perfectly over each other and switched from one to the next. I COULD TELL absolutely 0 DIFFERENCE.

So the question is...am I juts doing something wrong or is resize in ffdshow useless and a waste of CPU power?

Thanks,

Flaviu
Reply
Reply

AVS Top Picks

post #2 of 60
12/23/06 at 2:49pm
Thread Starter 
bump
Reply
Reply
post #3 of 60
12/23/06 at 7:10pm
does screenshot = screen capture or digital picture with camera?

what res is your display?

what res are you outputting to your display?

what setting in ffdshow? resize 2x, 3x? 2 "taps" or 4 "taps"?
Reply
Reply
post #4 of 60
12/23/06 at 7:20pm
Here's how I tested.

The best way to answer this is to compare frames.
Play a movie, pick a frame, pause MPC, open the file tab, Save image.
Open the saved pictures with "Windows picture & fax view" it's easy to switch between pictures and you'll be able to see any improvements.

You don't need to play the movie smoothly to tke the picture, so try out FFdshow now.
You just need to make sure you get the same frame.

I like screen shots that show rooms with a single person in it.
No certain colors, and I like the person to be still so there's no blur.

In Media Player Classic, under the Play tab. There's a frame step button. To use it, pause the dvd, and press the Frame step button, the movie will move one frame. This will help capture the exact frame.

With Seesaw. I made a picture with nothing but resize, so I could tell how sharp it was by default.

Then I made a picture with a Limitedsharpenfaster Ffdshow configuration.
Then I made a picture with a Seesaw configuration.

Finally, I used "Windows picture & fax view" to switch between frames and compare detail and noise reduction.
I usually just picked 2 frames, either LSF and Seesaw, or either Seesaw or LSf and the Plain resize picture to see a before and after.



The frame capture that doesn't use FFDshow resizing will be smaller than the one that does.
Now, press the + sign in the Windows picture & fax viewer to blow the smaller picture up to full screen.
It should look all blocky and ugly.

When your resizing the smaller picture to a larger picture using photoshop, photoshop is probably using Lanczos to resize it.

If you follow my instructions you can easily see the difference.
Reply
Reply
post #5 of 60
12/23/06 at 8:00pm
Thread Starter 
Quote:
Originally Posted by 8:13 View Post

Here's how I tested.

The best way to answer this is to compare frames.
Play a movie, pick a frame, pause MPC, open the file tab, Save image.
Open the saved pictures with "Windows picture & fax view" it's easy to switch between pictures and you'll be able to see any improvements.

You don't need to play the movie smoothly to tke the picture, so try out FFdshow now.
You just need to make sure you get the same frame.

I like screen shots that show rooms with a single person in it.
No certain colors, and I like the person to be still so there's no blur.

In Media Player Classic, under the Play tab. There's a frame step button. To use it, pause the dvd, and press the Frame step button, the movie will move one frame. This will help capture the exact frame.

With Seesaw. I made a picture with nothing but resize, so I could tell how sharp it was by default.

Then I made a picture with a Limitedsharpenfaster Ffdshow configuration.
Then I made a picture with a Seesaw configuration.

Finally, I used "Windows picture & fax view" to switch between frames and compare detail and noise reduction.
I usually just picked 2 frames, either LSF and Seesaw, or either Seesaw or LSf and the Plain resize picture to see a before and after.


The frame capture that doesn't use FFDshow resizing will be smaller than the one that does.
Now, press the + sign in the Windows picture & fax viewer to blow the smaller picture up to full screen.
It should look all blocky and ugly.

When your resizing the smaller picture to a larger picture using photoshop, photoshop is probably using Lanczos to resize it.

If you follow my instructions you can easily see the difference.


Umm, I did the exact same thing in the beginning regarding having the same frames, except I captured with "print screen" and then pasted in photoshop, BUT ...I was talking about ffdshow resize ALONE (no sharpening!) vs. MPC resize (to full screen, no ffdshow). So when I don't have ffdshow resize on, the 2 resultant screen captures are indeed the same size! There's no need to "blow things up" in fax viewer, which doesn't do a good job and why the default looks blocky...

Doing it my way, I could NOT see a difference between a resize done to fullscreen (1920x1200) by MPC vs. the 1920x1200 resize by ffdshow simply PLAYED in MPC...in full screen.

The only difference is with ffdshow resize, I use a lot more CPU!
Reply
Reply
post #6 of 60
12/23/06 at 8:04pm
Well I've tested just using MPC resize and apart from aspect ratio there was no difference from using Lanczos and Resize to not using FFDshow to resize.

I then tested resizing using FFDshow, and using Lanczos no taps, and 2 taps.
Again there was no difference.
I compared pictures like I've shown in my previous post.

I'll continue to use FFDshows resizing, but I'll set the taps to 0/default.
Reply
Reply
post #7 of 60
12/23/06 at 8:04pm
Thread Starter 
Quote:
Originally Posted by Fred Seger View Post

does screenshot = screen capture or digital picture with camera?

It should..., I would think.

Quote:
Originally Posted by Fred Seger View Post

what res is your display?

I tried this on my 1920x1200 LCD display.

Quote:
Originally Posted by Fred Seger View Post

what res are you outputting to your display?

1920x1200

Quote:
Originally Posted by Fred Seger View Post

what setting in ffdshow? resize 2x, 3x? 2 "taps" or 4 "taps"?

2x, lancosz
Reply
Reply
post #8 of 60
12/23/06 at 8:05pm
Thread Starter 
Quote:
Originally Posted by 8:13 View Post

Well I've tested just using MPC resize and apart from aspect ratio there was no difference from using Lanczos and Resize to not using FFDshow to resize.

I then tested resizing using FFDshow, and using Lanczos no taps, and 2 taps.
Again there was no difference.
I compared pictures like I've shown in my previous post.

I'll continue to use FFDshows resizing, but I'll set the taps to 0/default.

?? So are you saying there's no difference between using ffdshow with lanczos resize vs. mpc alone? ...what I said? What's the point of using it then?

I was hoping I was just doing something wrong...
Reply
Reply
post #9 of 60
12/23/06 at 8:06pm
Quote:
Originally Posted by Flav_cool View Post

Umm, I did the exact same thing in the beginning regarding having the same frames, except I captured with "print screen" and then pasted in photoshop, BUT ...I was talking about ffdshow resize ALONE (no sharpening!) vs. MPC resize (to full screen, no ffdshow). So when I don't have ffdshow resize on, the 2 resultant screen captures are indeed the same size! There's no need to "blow things up" in fax viewer, which doesn't do a good job and why the default looks blocky...

Doing it my way, I could NOT see a difference between a resize done to fullscreen (1920x1200) by MPC vs. the 1920x1200 resize by ffdshow simply PLAYED in MPC...in full screen.

The only difference is with ffdshow resize, I use a lot more CPU!

That's cool !
Thanks for that info. Maybe I won't use FFDshow to resize either, but I think I'm set in my ways.
Reply
Reply
post #10 of 60
12/23/06 at 8:28pm
I did that test the other day, and I couldn't tell a difference either; besides CPU difference. I was quite surprised at the results.
Reply
Reply
post #11 of 60
12/23/06 at 8:29pm
Thread Starter 
Quote:
Originally Posted by darkjedi664 View Post

I did that test the other day, and I couldn't tell a difference either; besides CPU difference. I was quite surprised at the results.

Could it be that MPC has started using lanczos by itself, without the need for the Lanczos resize filter in ffdshow?
Reply
Reply
post #12 of 60
12/23/06 at 8:29pm
Quote:
Originally Posted by Flav_cool View Post

?? So are you saying there's no difference between using ffdshow with lanczos resize vs. mpc alone? ...what I said? What's the point of using it then?

I was hoping I was just doing something wrong...

I don't know how your able to use prnt scrn. I can't. I can't do a proper comparison between just using MPC and MPC with FFDshow.
Reply
Reply
post #13 of 60
12/23/06 at 8:31pm
Thread Starter 
Quote:
Originally Posted by 8:13 View Post

I don't know how your able to use prnt scrn. I can't. I can't do a proper comparison between just using MPC and MPC with FFDshow.

You are most likely using overlay, and not VMR9 to render. You set this in MPC in the output setting.

Be aware that using VMR9 though (set in MPC), any output other than RGB32 (which you set in ffdshow) will show up as 16-235 colors when viewing!
Reply
Reply
post #14 of 60
12/23/06 at 10:04pm
VMR9 worked. I can now capture images.

Right click and save the images to your hard drive and compare them with Windows picture & fax viewer.

I see a difference.

No resize


Resized with FFDshow
Reply
Reply
post #15 of 60
12/23/06 at 10:06pm
Thread Starter 
^^ Yes, I see the difference in yours too. Are you sure you're not using a sharpen filter as well, with the ffdshow? That's what it looks like...

And are you outputting in RGB32, and if not, in what?
Reply
Reply
post #16 of 60
12/23/06 at 11:52pm
The resizing settings I used were.
Specify size: 1280x720
No aspect ratio correction
Lanczos
2 Taps

No other filter

Input all supported, output YV12
Reply
Reply
post #17 of 60
12/24/06 at 4:28am
Quote:
Originally Posted by Flav_cool View Post

^^ Yes, I see the difference in yours too. Are you sure you're not using a sharpen filter as well, with the ffdshow? That's what it looks like...

And are you outputting in RGB32, and if not, in what?

A lanczos resize algorithm has a sharpening effect regardless of whether you use an additional sharpening filter.
Reply
Reply
post #18 of 60
12/24/06 at 6:54am
A common problem with screengrabs is that they take individual fields rather than full frames, so you get half the vertical resolution. If anyone can be bothered, try downloading some progressive SD video and doing the MPC/ffdshow test with that.

Another issue might be that LCD screens won't show up sharpness so well, because they don't have good contrast ratios. FFDshow keeps the contrast nicely defined along the edges of screen items even after scaling, but on LCDs this contrast isn't there to start with.
Reply
Reply
post #19 of 60
12/24/06 at 8:53am
Thread Starter 
Quote:
Originally Posted by arfster View Post

A common problem with screengrabs is that they take individual fields rather than full frames, so you get half the vertical resolution. If anyone can be bothered, try downloading some progressive SD video and doing the MPC/ffdshow test with that.

Another issue might be that LCD screens won't show up sharpness so well, because they don't have good contrast ratios. FFDshow keeps the contrast nicely defined along the edges of screen items even after scaling, but on LCDs this contrast isn't there to start with.

Since I can see the difference in 8:13's pics, I think LCDs are just fine at seeing details...

Didn't know about the "field" problem, but it should catch, a deinterlaced, technically a progressive image.

I will try and do this test again myself...

But a question to 8:13, if you used YV12 output, how are you not seeing the washed out effect? Do you have the TV->PC conversion on in MPC, under internal filters, MPEG-2 filter?

EDIT: Ok I just redid the test and I fricken see the difference...ARRRGGGH.
Reply
Reply
post #20 of 60
12/24/06 at 10:21am
I didn't do anything with TV to PC conversion. I did calibrate my monitor though.
Reply
Reply
post #21 of 60
12/24/06 at 10:28am
Thread Starter 
Quote:
Originally Posted by 8:13 View Post

I didn't do anything with TV to PC conversion. I did calibrate my monitor though.

Yes, but the screenshot should still look like it's in 16-235 colors with YV12, and VMR9 on my monitor which isn't 16-235 color calibrated.
Reply
Reply
post #22 of 60
12/24/06 at 5:57pm
interesting
Reply
Reply
post #23 of 60
12/25/06 at 6:40am
Quote:
Originally Posted by 8:13 View Post

I'll set the taps to 0/default.

IIRC, the 'default' for lancos resize is 2 taps (don't think you can have zero).
Reply
Reply
post #24 of 60
12/25/06 at 9:09am
Are you using an nVidia card by any chance?

I fiddled with FFDShow for days - using all the recommended settings I could find - and never could get it to equal the quality of my Toshiba HD-A1 on std DVD. I turned off FFDShow and just started using the nVidia drivers and, viola, now the picture is pretty much spot on with the Tosh's scaling. And I didn't have to set a thousand parameters.

Just one man's experience...
Reply
Reply
post #25 of 60
12/25/06 at 11:11pm
You guys are doing something wrong. Every upscaling player/standalone scaler I've seen is less detailed than a properly setup htpc using ffdshow/lanczos. If you don't force lanczos to be used, the videocard uses its own bicubic scaling and that looks just plain mediocre.

I disagree with using 2x resize of native. Instead you should resize directly to the resolution of your display device. If you resize above what the videocard outputs, you are again using bicubic scaling to bring it back to the output res. My crt pj works best around 1280x720 so that's what I resize to. I tested using higher resize res's like 1440x960 and 1920x1080 and then having the videocard output 1280x720 but the image is always worse than just resizing directly to the target display res.

But some people will always go with the more is more mantra.
Reply
Reply
post #26 of 60
12/26/06 at 3:34am
Bicubic is a bit poor certainly, but downscaling is easy. To put it another way, you're scaling down what already exists, whereas with upscaling the machine has to guess what isn't there.

If you make a series of screenshots and flick between them, you can see that scaling to 3x DVD makes a difference, even on a 720p screen - nobody really understands why this is, and afaik it was discovered completely by accident :-)
Reply
Reply
post #27 of 60
12/26/06 at 6:22am
Quote:
Originally Posted by Jesse S View Post

You guys are doing something wrong. Every upscaling player/standalone scaler I've seen is less detailed than a properly setup htpc using ffdshow/lanczos. If you don't force lanczos to be used, the videocard uses its own bicubic scaling and that looks just plain mediocre.

....

Not if you've got one of the newer nVidia cards like I do. Its scaling is just plain better than what I could achieve with ffdshow.

I tried everything recommended everywhere I could find it. If I was doing something wrong, I don't know what it could be. After a few days someone tipped me off to the fact that, if I have an nVidia card, I should try the nVidia decoding and that did the trick.

One factor could be that I have a 1080p projector. Most who rant about ffdshow are using 720p. In any case, the nVidia processing is better than ffdshow (at least in the myriad of ways I tried it) and MUCH simpler. I have to say the A1 is just a smidge more natural in its picture, but the HTPC is close enough. Since my projector is 1080p I did not try scaling beyond that resolution.
Reply
Reply
post #28 of 60
12/26/06 at 7:44am
Quote:
Originally Posted by ebr View Post

Not if you've got one of the newer nVidia cards like I do.

As do I :-)

Incidentally, those who are most appreciateive about ffdshow scaling tend to be using projectors, while those least tend to be on <40" LCDs.

Not saying you're wrong btw, it may be different video cards have different performance, and you hit lucky. One of these days I'll do some screenshots of a bundle of different combinations of pictures/decoders/deinterlacers, post em somewhere without labels, and get people to rate them. Hopefully that might give a more objective perspective.
Reply
Reply
post #29 of 60
12/26/06 at 8:40am
Quote:
Originally Posted by arfster View Post

Incidentally, those who are most appreciateive about ffdshow scaling tend to be using projectors, while those least tend to be on <40" LCDs.

Or maybe those who would buy projectors are more knowledgeable on audio/video issues and have more persistence in tweaking their setups than those with <40-inch LCDs, and thus more likely to be able to configure FFDShow right.

Personally, I think I would prefer to take the dozens of hours that it would take to get FFDShow done right and use that time to watch dozens of movies instead.
Reply
Reply
post #30 of 60
12/26/06 at 11:15am
My HTPc using TT and FFDshow (lanczos x2 and a bit of luma sharpening to 1920x1080 card downscales to 1368x768 for output) is sharper and smoother than my A1 which whilst respectable exhibits ringing relative to the HTPC.

I also find hitting a high square pixel resolution and letting the card downscale is smoother with less aliasing than going straight to the output resolution in ffdshow.
Reply
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Home Theater Computers

AVS Top Picks

AVS › AVS Forum › Video Components › Home Theater Computers › FFdshow lancosz resize...no difference!?