|
|
| mariushudea |
| Posted: Jan 19 2008, 10:47 AM |
 |
|
Advanced Member
  
Group: Members
Posts: 54
Member No.: 22873
Joined: 19-January 08

|
When compressing a movie, the text in the Perf tab in the Compression status doesn't seem right, it appears like this:
Optimizing for Read mode Video streaming enabled Audio streaming
I also mentioned in the blog, if I have a sequence of images and at the open dialog I uncheck the Load linked segments, should Virtualdub still assume that I want all images loaded in sequence?
Create test video , RGB cube
Save as... XVID 800kbps chroma motion, high quality for b-frames, no audio ... at about 80% i checked Show decompressed output, the message that the display will appear at the next keyframe appeared, a keyframe passed at the output was not showed and after a few seconds Virtualdub crashed.
Probably related to the xvid patch mentioned a few pages before. Crashinfo.txt is here: http://savedonthe.net/download/11/crashinfo.txt
The second attempt checked show decompressed output right at the start, after the first keyframe it showed frames for a few seconds then crashed again, output is here:
http://savedonthe.net/download/12/crashinfo.txt 1.7.2 - 28312/release
The last frame of any video on my computer is always blue (the overlay color?), even in test videos... is something wrong in my settings or this is by design?
For example, with the test video Create chroma subsampling, the bar at the bottom starts at 0 and the preview image says Frame 0 but when i scroll to the end , the final frame is 1000 and it's blue, although the movie has in total 1000 frames ( from 0 to 999). The text below the bar says Frame 1000 (0:00:33:367)
If I use Lagarith Lossless codec, and check show decompressed output, the decompressed video never shows even though all frames are keyframes (it's because it's vfw codec and not direct show?). |
 |
| squid_80 |
| Posted: Jan 19 2008, 11:23 AM |
 |
|
Advanced Member
  
Group: Members
Posts: 594
Member No.: 13813
Joined: 22-January 05

|
Yep that's the show decompressed output vs. xvid crash. Perhaps you can confirm for Phaeron if it's fixed in this build since I've been slack and haven't followed up on my own bug report.
That last frame isn't actually there, it's just a placeholder for the end of video. Something about cuts being endpoint exclusive if I remember rightly. |
 |
| mariushudea |
| Posted: Jan 19 2008, 11:33 AM |
 |
|
Advanced Member
  
Group: Members
Posts: 54
Member No.: 22873
Joined: 19-January 08

|
Build 28322
Tried several times compressing with XVID the RGB cube.
Never crashed but also never showed the decompressed video output although there were several keyframes in the Video tab. Maximum i-frame interval is set to 250 in the xvid's configuration.
Correction: The video appears correctly after a keyframe is detected, but only if the "Show output video" is also checked. |
 |
| phaeron |
| Posted: Jan 19 2008, 09:28 PM |
 |
|

Virtualdub Developer
  
Group: Administrator
Posts: 7773
Member No.: 61
Joined: 30-July 02

|
Yes, the last frame is a one-after frame. If it weren't present, it would never be possible to include the last frame in a selection.
The fix in the test-4 build is working then? You do need "show output video" to be checked for decompression view to be visible, right now. |
 |
| mariushudea |
| Posted: Jan 21 2008, 08:23 PM |
 |
|
Advanced Member
  
Group: Members
Posts: 54
Member No.: 22873
Joined: 19-January 08

|
Yes, test-4 fix works.
I've stumbled across this site, it was submitted to Slashdot.
Out of curiosity, I tried resizing that picture of Dalai Lama from the page with Virtualdub using the Resize filter and indeed, the picture was all gray.
Does what this guy says apply to video as well? I'm not sure I really understood what he's talking about, I'd be really interested if you could give your feedback on this. |
 |
| phaeron |
| Posted: Jan 21 2008, 10:42 PM |
 |
|

Virtualdub Developer
  
Group: Administrator
Posts: 7773
Member No.: 61
Joined: 30-July 02

|
Yes, it does apply, and yes, VirtualDub does suffer from the same issue. However, there are a few points I should make on this subject.
First, the effect is much more subtle on actual imagery. The example given is an extreme case since it consists of an alternating checkerboard. It can be a problem on any sort of video, but I doubt you'd see the difference in practice, and there are much bigger transgressions generally involved in video processing, such as black/white levels and sloppy rounding.
Second, the error depends on the color space used to encode the video. If you're really concerned about linear vs. gamma space you would not want the resampler to be handling this problem -- you would want to convert from gamma space into linear space at the start of the rendering pipeline and keep it there until the very end. All linear filters will exhibit this issue, including blurs and sharpens.
Third -- and this is the ugly problem -- 8-bit per channel is nowhere near enough precision to handle conversions between linear space and gamma space. Jim Blinn has a nice article in his Dirty Pixels book where he shows that in order to maintain full 8-bit precision in gamma=2.0 space you need 17 bits in linear space. To properly switch to a linear space you thus need at least 96 bits/pixel. This is beyond the ability of most fast video processing libraries and imposes serious performance penalties.
Bilinear vs. bicubic resampling is a similar issue, really. Mathematically bicubic is far superior, and there are corner cases where bicubic is visually superior, such as gradient determination. For most video, though, the visual difference between bilinear and bicubic is minor and nowhere near that of bilinear vs. point sampling. |
 |
| phaeron |
| Posted: Jan 27 2008, 04:03 AM |
 |
|

Virtualdub Developer
  
Group: Administrator
Posts: 7773
Member No.: 61
Joined: 30-July 02

|
New version posted:
http://www.virtualdub.org/beta/VirtualDub-...1.8.0-test1.zip http://www.virtualdub.org/beta/VirtualDub-...test1-AMD64.zip http://www.virtualdub.org/beta/VirtualDub-....0-test1-src.7z
This is the first release candidate for 1.8.0/experimental.
Changes in this version:
- Merged in fixes from 1.7.8 (which isn't out yet, but will probably go out soon).
- I fixed a nasty bug where doing edits in direct mode and saving only part of the timeline via selection could result in excess frames pushed back in by the rendering engine. (Basically, it's using the wrong frame number to determine where frames have to be added back in, so it adds in more than are necessary to satisfy key frame restrictions.) I considered fixing this for 1.7.8, but punted because it became too risky and it's been in there for a while, so the fix will go out first in 1.8.0.
- Fixed some bugs regarding preview and filters that change frame rates.
- Merged ela doubler and ela doubler 2 filters into bob doubler and added config settings for field polarity.
- Filter cropping now has fast and precise modes. Fast mode has the same behavior as the previous test build in that it will round off the cropping bounds depending on the incoming format; precise mode will instead automatically perform a format conversion as necessary, always doing the exact crop requested. For instance, if the incoming video is YV12, the filter engine may convert it to either YV16 or YV24 in order to do the crop. The default is precise mode.
- squid_80's direct mode null frame crash is fixed.
- Fixed slight delays when starting renders in command-line mode.
All feedback welcome.
One bug that isn't fixed in this version is that if you have a filter that changes the frame count, input preview doesn't work. In fact, any video mode other than full processing mode may reference the wrong frames. The reason is that I painted myself into a corner by making the timeline use post-filtered frames, which is incorrect when a filter is altering the frame translation. I thought long and hard about this and came to the conclusion that post-filtered frames are the way to go, because otherwise frame-doubling filters are annoying to use, but I need to do enough work to fix this problem that I think it'll have to wait until 1.8.1. I need to rewrite the IVTC engine too because it also doesn't interact well with frame doubling, although it needs a lot of help in general, and realistically I care a lot more about field doubling working than IVTC working. |
 |
| Barough |
| Posted: Jan 27 2008, 05:29 AM |
 |
|

Advanced Member
  
Group: Members
Posts: 95
Member No.: 22677
Joined: 24-December 07

|
Thnx for the new v1.8.0 Test 1, will take a look @ it during the coming week
--------------------
 |
 |
| olnima |
| Posted: Jan 29 2008, 07:37 AM |
 |
|
Advanced Member
  
Group: Members
Posts: 204
Member No.: 17204
Joined: 12-November 05

|
Hi phaeron, a few versions ago You solved a bug which has made it again into new 1.8test1:
if I want to fade-in a captured video by using the curve-editor and "fill" - filter, everything is ok. But if I cut the beginning afterwards, I also cut away my fading.
In the old thread about this http://forums.virtualdub.org/index.php?act...hl=entry55732
this was your answer: < Looking at the code, though, I think I know what's going on. Did you delete frames before the start of the video? I think the code is using the source frame instead of the timeline frame to check the filters, and if you shift the frames that'll break the check. >
//edit1: second bug is the following: if i save the processing-settings with filter "fill" and "blend" and two points (in and out for fading), "blend" and the two points are ignored when re-importing these saved-processing settings.
example: > VirtualDub.video.filters.Clear(); VirtualDub.video.filters.Add("fill"); VirtualDub.video.filters.instance[0].Config(0,0,0,0,0x000000); declare curve = VirtualDub.video.filters.instance[0].AddOpacityCurve(); curve.AddPoint(0, 1, 1); curve.AddPoint(50, 0, 0); VirtualDub.audio.filters.Clear(); < this is saved correctly (don't know if it is correct but saving vcf-file results in the same code as in older versions), but if I re-import it, the filter-setting "blend" and the points are lost (ignored).
Hope You know what I mean Olnima |
 |
| phaeron |
| Posted: Jan 30 2008, 07:14 AM |
 |
|

Virtualdub Developer
  
Group: Administrator
Posts: 7773
Member No.: 61
Joined: 30-July 02

|
Nice catches. Curve bug was the same, but the underlying cause was different.
http://www.virtualdub.org/beta/VirtualDub-...1.8.0-test2.zip http://www.virtualdub.org/beta/VirtualDub-...test2-AMD64.zip http://www.virtualdub.org/beta/VirtualDub-....0-test2-src.7z
|
 |
| olnima |
| Posted: Jan 30 2008, 08:11 AM |
 |
|
Advanced Member
  
Group: Members
Posts: 204
Member No.: 17204
Joined: 12-November 05

|
...nice fixes , thank You, seems to work - means that I will post here if it doesn't.
Btw: Any chance to implement an option to create a vcf-file for making a fade-in at the beginning AND a fade-out always at the end of the avi without knowing its total number of frames (after cutting)?
Thanks again, Olnima |
 |
| mariushudea |
| Posted: Jan 30 2008, 05:45 PM |
 |
|
Advanced Member
  
Group: Members
Posts: 54
Member No.: 22873
Joined: 19-January 08

|
In Virtualdub 1.7.7, if you check Show output video and Show decompressed output and the codec can not decode content (x264 encoder for example), then the proper message appears :
Unable to display compressed video: no decompressor is available to decode the compressed video
The problem is that after this happens, the output video never shows, no matter what checkbox I select. I should see the output video if I check only "Show output video" |
 |
| phaeron |
| Posted: Jan 31 2008, 09:08 AM |
 |
|

Virtualdub Developer
  
Group: Administrator
Posts: 7773
Member No.: 61
Joined: 30-July 02

|
@mariushudea: Thanks, bug fixed. You'll see the fix in the next test release if I push one out.
@olnima: Unlikely, I'm afraid. The reason is that I've found that doing end-relative things like that are bug fountains -- basically I end up having to write a bunch of code to keep updating the absolute position from the relative position, which invariably goes wrong. The end-of-selection was an example, which was a constant headache.
What you're asking for can be done right now by editing the script directly, using VirtualDub.video.length -- it's just that I don't think I'll be creating a UI for that. |
 |
| olnima |
| Posted: Jan 31 2008, 09:27 AM |
 |
|
Advanced Member
  
Group: Members
Posts: 204
Member No.: 17204
Joined: 12-November 05

|
Thanks for the suggestion, editing vcf-file manually (without having an ui) would be more than enough. Problem is, if I load video and vcf-file via commandline, I will loose end-points after cutting. After playing around a bit I found out, that "VirtualDub.video.length" also holds the number of source-frames instead of the timeline-frames. Can You change this behavior too analog to the upper mentioned and solved "curve-bug"? With this I could load a 2nd vcf-file after cutting, only for fading-out at the end.
Olnima |
 |
| phaeron |
| Posted: Feb 1 2008, 07:40 AM |
 |
|

Virtualdub Developer
  
Group: Administrator
Posts: 7773
Member No.: 61
Joined: 30-July 02

|
Can't do that, because it would break backwards compatibility. That's what VirtualDub.subset.length is for, though. |
 |