Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion. Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules. |
21st February 2015, 16:10 | #41 | Link | |
Registered User
Join Date: Dec 2013
Posts: 753
|
Quote:
|
|
21st February 2015, 16:19 | #42 | Link | |
Registered User
Join Date: Oct 2012
Posts: 7,926
|
Quote:
you could add IVTC to deinterlacing. and you my add an comment that force film mode doesn't work with native DXVA |
|
21st February 2015, 16:38 | #43 | Link | ||
Registered User
Join Date: Sep 2004
Posts: 146
|
Quote:
Quote:
|
||
21st February 2015, 19:32 | #44 | Link | |
Registered User
Join Date: Jul 2014
Location: Las Vegas
Posts: 177
|
Quote:
Also in games for example, 4xSSAA is 2x in the horizontal and 2x in the vertical. But it is still called 4x, because the result is 4x greater. Well power of 2 can mean anything. Maybe 2^2 (4x) is reffered to as doubling and 2^4 (16x) is quadrupling, based on the amount of pixels they give. |
|
21st February 2015, 20:42 | #45 | Link |
Registered User
Join Date: Dec 2008
Posts: 496
|
No need to make this so complicated.
"Image doubling" doubles horizontal as well as vertical (width and height) pixels, while quadrubling is basically 2x "Image doubling" applied. And since this is directly connected to the way NNEDI3 works, you can 2x2x2x2.... as much as you would like with subsequent "doubles" (at least theoretically). But then again, even most of the current GPUs struggle with doubling (with at least NNEDI3 32 neurons). So, basically "to be continued"? madshi always uses the "makes sense, why make it anymore complicated than it already is" approach, there is no rocket-science behind this, even though madshi usually comes up with pretty great ideas and surprises us. Last edited by iSunrise; 21st February 2015 at 20:52. |
21st February 2015, 22:18 | #46 | Link |
Registered User
Join Date: Dec 2014
Posts: 1,127
|
Your chart appears logical to me. But I also question why the output would be converted to RGB, immediately converted back to YCbCr only to be converted back to RGB once again when image doubling. Are you sure this is correct?
|
21st February 2015, 23:34 | #48 | Link | |
/人 ◕ ‿‿ ◕ 人\
Join Date: May 2011
Location: Russia
Posts: 643
|
I remember reading about it too. Probably keeping rgb between shaders is easier (from development point of view), even though it's not effective.
Quote:
NNEDI is a 1-dimensional filter that doubles the resoulution. In case on anamorphic video it's possible that it will run it in only one direction (e.g. 16:9 1440x1080 with doubling set to always). In other cases it runs it twice (1 in each direction). So you should think not about resolution, but about directions. It's just that keeping separate settings for 2 directions doesn't make much sence, so they're combined. About debanding. RGB: if you check shader code, you'll see that it converts from rgb to ycbcr and then from ycbcr to rgb. Subsampling: I believe that DX9 doesn't support writing into 16-bit subsampled textures, so... Emulating it with 2 textures is not only painful, but also will make it slower (+25% pixels to write, doesn't matter that they have less channels). Actually I don't believe that it even supports subsampled 10/16-bit textures (this is probably the reason why we won't see 10-bit native DXVA in madVR). So doing it after chroma upsampling is hell lot easier and doesn't impact performance. Last edited by vivan; 21st February 2015 at 23:43. |
|
21st February 2015, 23:57 | #49 | Link | |||||||||
Registered User
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,407
|
Quote:
Quote:
Using your nomenclature 2x the resolution implies an irrational scaling factor of square root of 2 or only doubling one dimension. It sounds great for people trying to sell new models of digital cameras but no one else talks about resolution this way. I am describing a standard so everyone uses the same words to mean the same thing. madshi obviously agrees given image doubling and as this thread is about madVR options let us stick to the terminology madVR uses. Everyone else who works with digital image processing thinks madshi got it right so this should not cause confusion. When you scale an image in Photoshop to "200%" you get 4 times the number of pixels. Quote:
I have not been able to find a statement from AMD or Nvidia but this is representative of what I was able to find: Quote:
Quote:
Quote:
Quote:
Quote:
Last edited by Asmodian; 22nd February 2015 at 00:43. |
|||||||||
22nd February 2015, 01:44 | #50 | Link | |
Registered User
Join Date: Jul 2014
Location: Las Vegas
Posts: 177
|
Quote:
You are correct that we are talking about spacial resolution. If an image is 3840x2160, but its spacial resolution is lower, meaning the pixels are much more spread out, it might look the same as the same image at 1920x1080 at the correct distance. But, the 3840x2160 image still has more information, it just displayed incorrectly/differently. Photoshop refers to 200% as 4x the pixels because it scales the horizontal and vertical together. Just like nnedi3. If it only scaled in one dimension, what would you call it? What is 3840x1080 compared to 1920x1080... using your defintion? You can't describe it. But you can say it is 2x the resolution, and 3840x2160 is 4x the resolution. It makes sense to refer to "scaling" as 2x, because you scale both directions together. But it does not make sense to refer to the entire resolution as 2x. In triple A games, 4xSSAA is 4x the resolution. If 4xSSAA was 16x the pixels, it wouldn't exist in any games, you would need 4 gtx 980s running at 2200mhz cooled by ln2, obviously not realistic to implement. Also, check the dynamic super resolution in your nvidia control panel. I am currently on a 1920x1200 monitor, and 4x the resolution is referred to as 3840x2400. Same multiplier scale is used for all the other dynamic super resolutions. 2x the resolution is 2715x1697. Picture here: http://i.imgur.com/ID3XoBc.png Questionable. Nvidia inspector, as shown in my next post, allows you to configure the horizontal and vertical independantly of each other. This can provide better image quality without destroying your frame rate, as the picture rendered at the higher resolution is downsampled. Obviously this might not work properly in some games, or madvr for that matter, which is why it isn't an option in the applications itself, but rather the drivers. A 1920x1080 image scaled by 4x, is now 400% larger. Photoshop refers to it as 200% because Photoshop is an image editing tool. You can select to only the scale the horizontal or vertical by 200%, doesn't mean the amount of pixels increased fourfold. The 200% that results in 4x the pixels just means you increased the resolution by 2 in both directions. Last edited by Arm3nian; 22nd February 2015 at 02:37. |
|
22nd February 2015, 02:07 | #51 | Link |
Registered User
Join Date: Jul 2014
Location: Las Vegas
Posts: 177
|
Also check this: http://i.imgur.com/7VaN0Tc.png
Notice how SSAA is referred in both the horizontal and vertical resolution. 2*2=4, so 2x2SSAA is shortened as 4xSSAA. This can be confirmed by looking below at the combined SSAA and MSAA options. 32xS is 2x2SSAA + 8xMSAA, 2*2=4, 4*8=32. This is the most logical method of labeling the values. Last edited by Arm3nian; 22nd February 2015 at 02:08. Reason: big picture sorry |
22nd February 2015, 02:39 | #52 | Link | |
Registered User
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,407
|
Quote:
To be precise "2x the resolution" is ambiguous, it could have either interpretation. It would be more precise to say "2x the resolution horizontally and vertically" or "twice the number of pixels". There are reasonable reasons to talk about resolution as the total number of pixels but that is not the convention normally used when discussing digital images and video (unless you are a marketing department). 2x scaling gives 2x the resolution (Nvidia's DSR notwithstanding, they only want to describe the performance hit and do not care about conventions). Anyway I am happy to agree to disagree, only know if I say "2x the resolution" I mean 4 times the number of pixels. |
|
22nd February 2015, 03:33 | #53 | Link | |
Registered User
Join Date: Jul 2014
Location: Las Vegas
Posts: 177
|
Quote:
|
|
22nd February 2015, 11:04 | #54 | Link | ||
Registered User
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,407
|
Quote:
Quote:
If you have twice the resolution you expect to be able to resolve details at half the size. For a 2D space you need four times the number of pixels to resolve details at half the size. e.g. If the smallest filament you can resolve with an imaging system is 0.1 mm in diameter and you need to be able to resolve one with a diameter of 0.05 mm you need a sensor with double the resolution. It is not useful to raise the change in resolution required to the power of the number of possible axes when discussing a needed change in sampling. It simply removes meaning. Saying you need 8 times the resolution to resolve a 3D object at half the size is pointless, why not say you need 8 times the samples if that is how you want to express it? Resolution is a term used to describe data sets which are both a type of information themselves and contain information. The relative change in resolution equals the relative change in the number of samples only when talking about a data set with one dimension. At this point I think I have expressed my option on the matter as well as I can. |
||
22nd February 2015, 11:30 | #55 | Link | |
Registered User
Join Date: Jul 2014
Location: Las Vegas
Posts: 177
|
Quote:
Nvidia most likely uses 4x in the DSR to show performance impacts, as the gpu will render 4x the amount of pixels. For photography and video I see why 2x would make sense. 4x when talking about scaling might give the wrong impretion on expected quality. The size of the image is important to consider. It would be to the benefit of all if everyone was more descriptive, but as you already know, it is hard to get an entire industry to agree on something. |
|
24th February 2015, 03:02 | #57 | Link | |
Registered User
Join Date: Dec 2014
Posts: 1,127
|
Quote:
A 720p source has 2.25x the number of pixels of a 1080p source. However, the actual increase in vertical resolution or (pixels per inch) is only 1.5x. The amount of dots per inch is a better indication of quality than the total number of pixels. Moving from 1080p to 4K, the number of pixels increases to 4x, but the scaling factor only improves to a mere 2x. A better question is whether anyone can resolve the extra pixels per inch from their seating distance, which they probably cannot. |
|
24th February 2015, 05:36 | #58 | Link | |
Registered User
Join Date: Jul 2014
Location: Las Vegas
Posts: 177
|
Quote:
Think about a water bottle. If I have a water bottle that is 1ft tall, and I increase it to 2ft, I now have a water bottle that is 2x bigger, because it holds 2x the water. If I increase it to 4ft, then I have a watter bottle that holds 4x as much as water as the initial bottle, making it 4x bigger. Now lets say I invent a gun that makes objects 2x bigger and I shoot my water bottle. Is it going to make it only 2x taller? No, it is going to multiply ALL dimensions by a factor of 2. Now the water bottle does not hold 2x as much water, it holds 8x as much water. So in theory, increasing something by a factor of 2 is relative on the amount of dimensions. In 1d, a factor of 2 increases the size by 2, in 2d, a factor of 2 increases the size by 4, and in 3d, a factor of 2 increases the size by 8. In other words, it more precise to refer to these things by what matters. In resolution, the pixel count, which directly relates to the performance. And in the case of the water bottle, to the volume, which directly reflects the amount of water it can hold. Making something 2x bigger just implies that the actual value did not increase linearly, so why not just refer to the actual value directly... makes it more simple, and more logical. Resolution is just another word for pixel count. A resolution of 2 means 2 pixels. A resolution of 3840x2160 means 8294400 pixels. The image is just the object. The resolution is the quantity being defined. Just like a water bottle is the object and the volume is being defined. This is why madshi refers to it as image doubling, because making an image 2x bigger creates 4x the pixels. Saying the resolution in 1920x1080 to 3840x2160 doubled makes no sense, because it quadrupled. The image itself doubled. 4k might provide an increase in perceived image quality on a TV, but it is more noticeable on a monitor to me since I sit close. Last edited by Arm3nian; 26th February 2015 at 08:41. |
|
26th February 2015, 03:52 | #60 | Link | |
Registered User
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,407
|
Quote:
Thanks, I am glad it helped a bit. Last edited by Asmodian; 26th February 2015 at 04:02. |
|
|
|