Jump to content

720p V 1080i: Whats Better For Fast-action Sport?


Recommended Posts

Owen,

How are you accessing hardware deinterlacing for the 8800 ?

With the 8600GTS I have not been able to surpass the video performance of the Sony HD-STB nor have I found any deinterlacing options within the Nvidia software for the 8x00 cards in either XP or Vista.

I use DNTV Live and either the nVidia decoder or the Cyberlink decoder that came with Power DVD 7.3, both with hardware assistance (DXVA) enabled. Neither are perfect as they occasionally loose the plot on some content.

I really don’t watch free to air TV. The Sony STB and the tuner in my HTPC have seen maybe two hours use between them in the last 4 months.

I did watch some motor racing on channel 7 (SD) about 3 weeks ago, and was swapping between the PC and the STB to see how they compared. Overall picture quality is very similar, but the horizontal scrolling text (ticker) on the bottom of the screen was noticeably better on the HTPC. The text was clean and free of flicker, where as via the STB it was ragged and flickery. The deinterlacer in the SXRD may be better then the one in the STB, but unless we change the output format on the STB to watch 576i or 1080i we have to live with what the STB does with 576i.

When I first installed the STB I spend an hour or so viewing 1080 HD loops and comparing it to the PC. Even though the STB is connected via the $2 Component cables that came with it, there was only a subtle difference between it and the HTPC connected via HDMI at 1080p. The STB is running 1080i output so that the DRC settings can be used on the SXRD. The DRC can be adjusted to give a sharper picture then is possible without it, but sharper is not better and the picture was cleaner via the PC. I generally find the picture better with DRC disabled.

I have no doubt that a HDMI cable would provide a cleaner image from the STB then a $2 Component cable, but I cant be bothered to get one just to watch a few hours of FTA TV in a year.

A PC is not the ideal device to watch live FTA TV IMHO as it’s not as consistent as a STB, although they can work fine. The family 86cm CRT runs digital free to air from a HTPC all the time (no STB) and gets heaps of problem free use.

For recorded content I wont use anything but a HTPC. Foxtel-Austar is also better when processed via the PC.

Link to post
Share on other sites
  • Replies 131
  • Created
  • Last Reply

Top Posters In This Topic

Thanks Owen for the reply.

With regard to DXVA being enabled with either the nvidia or Cyberlink powerDVD 7.3 decoder, have you been able to confirm whether or not hardware acceleration is actually engaged and is there any difference in picture quality if DXVA is not enabled ?

I have no doubt that a HDMI cable would provide a cleaner image from the STB then a $2 Component cable, but I can't be bothered to get one just to watch a few hours of FTA TV in a year.

I have the HD-STB connected to the SXRD via HDMI.

Link to post
Share on other sites
Hey Owen drsmith, how did the Channel 7 footy (1080i) look on Sunday afternoon on your SXRD??

I didn't watch it. I thought only 10 was broadcasting one game per week in HD. I can't see any mention of 7 broadcasting AFL in HD here.

I use DNTV Live and either the nVidia decoder or the Cyberlink decoder that came with Power DVD 7.3, both with hardware assistance (DXVA) enabled. Neither are perfect as they occasionally loose the plot on some content.

I have done some experimentation with DNTVLive! in XP with a segment of tennis recorded about six months ago.

With the Cyberlink Power DVD 7.3 decoder there is definately an improvement in picture quality with DxVA enabled.

Link to post
Share on other sites


Anyone wishing to experiment may find the following useful:

From this pdf:

The SVT High Definition Multi Format Test Set Version 1.0 Feb.2006

Sveriges Television www.svt.se Lars Haglund 1

The SVT High Definition Multi Format Test Set

1. Introduction

This Document contains information about the
SVT High Definition Multi Format Test Set

picked from the demanding, but not unduly so, multi-genre TV-program ‘Fairytale’ (by SVT)

mastered in 3840x2160p/50.

All sequences from this set may be distributed freely (complete or in parts) as long as the

copyright and the restrictions of the use of the sequences are not violated.

This document has to be distributed with any part of the test set that is distributed.

It is not allowed to charge a fee for the distribution of the sequences. Anybody receiving

material from this data set is kindly asked to further distribute these sequences to other parties

to reach a super distribution effect.

All sequences from this set were filmed in 50 fps with professional, high end 65mm film

equipment by SVT in October 2004. The utmost care has been taken when converting the

film into digital data, the same applies for all conversion steps applied afterwards.

The data itself comes in various different resolutions which are widely used – all in 50 Hz

motion portrayal. Lower resolutions were gained by filtering the 2160p/50 master. At all

resolutions the bit depth is full interval 16 bits per (RGB) colour plane to obtain the very high

quality of the original shots.

Details concerning shooting of the sequences and post-processing steps can be found in the

section ‘Technical Information’.

The test sequences are downloadable from here. The explanatory material states: "These sequences may only be used for the purpose of developing, testing and presenting technology standards." Available formats include 1280x720 p50 and 1920x1080 p50, and 1920x1080 i25*, all derived from the original 3840x2160 p50.

Note the format is of individual frames in sgi format. A freeware individual frame viewer and converter (XnView v1.91.1) can be downloaded from here.

Each frame of the 500 frames (numbered 07111 to 07610) from the 1080p ftp download site is 12.4MB, so the total download would be about 6.25GB. The 720p frames are about 5.53MB each, i.e. about 2.8GB for the 500 frames.

It could be someone has already downloaded these, assembled them into a movie, compressed the video, and uploaded the result somewhere.

*i25 as used here means 50 fields per second, resulting in 25 interleaved frames. Each frame available for download is therefore a weave of two fields (even and odd lines) captured 1/50th of a second apart in time

I have made up some very short videos using frames from the SVT High Definition Multi Format Test Set above. They are a means of "presenting technology standards" but are for use at own risk. Although I have taken care, I cannot guarantee the technical standard of the videos.

They are as follows:

Clip1: CrowdRun720p50.avi 19.23MB 20Mbps

Clip2: CrowdRun1080i25.avi 24.12MB 25Mbps

Clip3: CrowdRun1080p50.avi 31.8MB 33Mbps

As web bandwith is limited, I'd ask that if you download you please save, so that you do not need to download the same file more than once.

They are each 10 seconds long, and show exactly the same picture content -- a crowd of people running. It appears to be a cross-country run, with a very large number of participants. This represents a challenge for any video compression algorithm.

Rather than doing my own resizing or conversions from progressive to interlaced, I simply downloaded the relevant source frames already of the apropriate format, converted them to bmp images, loaded them into VirtualDub, and saved them with ffdshowddshowfilter set for H.264 compression (one pass - 85% quality).

I originally tried MPEG2 compression but the file sizes became very high for reasonable quality. I found the H.264 encoding was effective with the progressive material, even at low bitrates. (In contrast, the interlaced material needed a certain minimum bitrate or it would become "sketchy" and difficult to watch.)

Comparing the visual quality

  • There seems no doubt clip 3 is the best; giving smoothness and detail.
  • Clip 1 is easy to watch, being quite smooth.
  • Clip 2 has a certain disjointedness [the exact appearance will depend on the characteristics of the de-interlacing for the viewing device], but could be said to have more detail than clip 1.

I think I am more inclined towards clip 1 than clip 2, despite my usual preference for 1080i, with less fast moving material.

I have not found much true 1080p material on the web. Although called 1080p it is usually from 24fps [or 25fps] film.

It's a pity, because true 1080 p50 can look very impresssive.

Link to post
Share on other sites

It’s a shame you have invalidated your test by using un optimised h.264 compression.

The video should have been left in its original form, or at the very least compressed with optimised Mpeg2, which is what we use for TV transmission.

Link to post
Share on other sites
It’s a shame you have invalidated your test by using un optimised h.264 compression.

The video should have been left in its original form, or at the very least compressed with optimised Mpeg2, which is what we use for TV transmission.

The original form (bmp) for the 1080p immediately prior to compression was 2.90GB, for the 1080i it was 1.45GB, and for the 720p it was 1.29GB.

The compression factor used was more in line with HD-DVD than with free to air.

There was little point in multiple pass encoding, as the picture content was highly changeable through the whole of the 10 seconds of the clip.

[1080p of course is not an existing FTA or high definition disk (HD-DVD or Bluray) format.]

I found that lower compression factors were a little less kind for the 1080i than the 720p or 1080p. [both bottom field first and top field first options for the interlaced format were tried.]

In addition to Mpeg 2 and H.264, a DivX codec was tried.

To my eyes, the three picture formats preserved their different characters, irrespective of the compression codec.

I was more interested in the basic characteristics of the formats than how curent bit-rate limitations of FTA further reduce the quality.

With no compression, the 1080i avi will play on Zoomplayer only at a reduced speed, even with clip size reduced to 50 frames (2 seconds' worth). I may experiment further with playing the frames uncompressed, or with very mild compression, and report the result.

Link to post
Share on other sites

Try watching uncompressed video at full speed for say 10 seconds to get a feel for how it performs. If your system cant cope, you cant do a valid comparison.

The problem will be getting good hardware deinterlacing working with uncompressed content. PC replay systems are not optimised for uncompressed video, and ineffective deinterlacing will ruin your comparison.

h.264 or the PC playbacks systems for it have big problems with 1080i in my experience. 1080i in Mpeg2 works faultlessly on the same system because PC systems are now well optimised to handle 1080i in Mpeg2 format.

If you want to create 1080i in Mpeg2 you will need a professional grade compression codec to do it justice.

I’ll download the original source video and see if I can play it properly on my system if you like. I’m very used to watching 1080i content and I’ll know if it does not look the way it should.

Link to post
Share on other sites


Owen thanks for your comments.

The download of the 1080i frames is about 3GB (250 frames at 12Mb each in SGI format). The five hundred 2160p frames come in at about 23GB :blink: if they are needed. [They show film grain and are probaby not the best as 2160p source material as such, but might be useful for deriving a lower resolution for test purposes, such as 576p.]

As a work around for the uncompressed playback issue, I've been able to play cropped versions of the 1080i and 1080p avi's I created, at full speed. The cropping was to remove 600 pixels from the left, 600 from the right, and 504 from the top, thus reducing the 1080i or p to 576i or p. The resulting 576p played superbly.

There was still the issue of deinterlacing with the resulting 576i. I used the Smooth Deinterlacer (Gunnar Thalin) filter with VirtualDub* and the output was indeed very smooth, and easy on the eye.

I then compared the uncompressed 576p (cropped from the uncompressed 1080p) with the uncompressed 576i (deinterlaced from the cropped 1080i). Both were very smooth, but the 576p had noticeably more detail.

The quest

I would observe that there is no reason why the deinterlacing that a graphics card does cannot be emulated in software, even if the emulation is much slower.

We seem to be on a quest to find a deinterlacing system that provides smooth motion and detail, in circumstances where only 540 lines are refreshed each 1/50th of a second, and a lot of the image detail changes appreciably each 1/50th of a second; in comparison with a system where 720 lines are refreshed each 1/50th of a second.

It is often stated that 720p is better for fast-action sport than 1080i. At this stage I have not seen evidence to contradict that position.

* An Avisynth File was needed to load in the uncompressed 576i, as follows:

AVISource("576isource1.avi")

AssumeFrameBased()

ComplementParity()

SeparateFields()

This split the 250 interlaced frames into 500 half-height 576i frames; which the Smooth Deinterlacer filter assembled into 500 full height progressive frames.

Link to post
Share on other sites
  • 3 months later...

Owen has recently started a new thread: Is 1080p Superior To 1080i ?, The answer may not be what you expect. That is prompting me to provide some further video samples. They tend to confirm my impressions of a few months ago.

The first two samples use the 1080p still frames from the Crowd Run sequence. (See post #125 above for a description of where this material originated). And the third sample uses 720p still frames from the same sequence.

The steps to create the raw frames were as follows:

1080i50

Load the 500 1080p50 frames [in bmp format, converted from sgi] into Virtualdub using an AviSynth script that creates interlaced frames. Use the Virtualdub level filter to map video levels 0-255 to 16-235, and save as an uncompressed AVI. Without the mapping, black detail would be lost in the later Mpeg2 encoding.

1080p25

Load the 500 1080p50 frames into Virtualdub. Save them as an uncompressed AVI, discarding each second frame, and use the level filter to map the intensity levels.

720p

Load the 500 720p50 frames into Virtualdub. Save them as an uncompressed AVI after setting the level filter.

It was then possible using a trial version of Main Concept Pro to encode the raw files as interlaced (upper field first) or progressive as the case required, using a data rate of approximately 25Mbps, which is a typical rate for encoding Blu-ray disks.

The results (each file is about 35MB) are:

1. CrowdRun25Mbps-1080i50.m2v downloadable from
.

2. CrowdRun25Mbps-1080p25.m2v downloadable from
.

3. CrowdRun25Mbps--720p50.m2v downloadable from
.

As download bandwidth is strictly limited, I'd ask that anyone performing a download makes a point of saving the file, so as to avoid any need to download a particular file more than once. [The bandwidth resets at the end of each month.]

It's interesting that the smoothness of the 720p50 makes the runners appear not to be running as fast.

I suggest this old thread might be allowed to lapse, unless someone has specific comments about 720p vs 1080i.

Edited by MLXXX
Link to post
Share on other sites

What a joke, that looks like one field of 1080i video (540 lines), or simple bob deinterlacing.

The guy needs to re-evaluate his image capture methods, no way 1080i ever looks like that when displayed properly.

If I ever saw anything like that on my display I would be looking for what the hell was wrong with the replay system.

Link to post
Share on other sites


Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Recently Browsing   0 members

    No registered users viewing this page.




×
×
  • Create New...