Converting AVI (DV) file with changing PAR (ffmpeg?)

[rgr]

I have an AVI file ripped from a miniDV tape (PAL).

The aspect ratio of the file changes from 4:3 to 16:9 every now and then.

MPC-HC plays such a file correctly, but AviSynth assigns DAR from the first shot to the entire file.
Can Avisynth even read such files correctly and "transfer" such a frame flag to ffmpeg (with -avisynth_flags all)?

I use ffms2 to open the file (if that matters).

[poisondeathray]

I have an AVI file ripped from a miniDV tape (PAL).

The aspect ratio of the file changes from 4:3 to 16:9 every now and then.

MPC-HC plays such a file correctly, but AviSynth assigns DAR from the first shot to the entire file.
Can Avisynth even read such files correctly and "transfer" such a frame flag to ffmpeg (with -avisynth_flags all)?

I use ffms2 to open the file (if that matters).

Avisynth should be able to parse that information with frame properties if you use a source filter that reads it (e.g. ffms2, as you are using) - you can check with propshow() and _SARDen, _SARNum are read. Check to see if it switches on your video


But ffmpeg avs demuxer does not pass that information. Currently it passes

_FieldBased 
_ChromaLocation
_Primaries
_Transfer
_Matrix
_ColorRange

You can make a request for someone to update the ffmpeg avs demuxer with more frame properties

[rgr]

ffms2 doesn't show correct _SarNum/Den (probably older version, but new ones have problems with file names). LWLibavVideoSource does.

But ffmpeg with "-avisynth_flags all" can read DAR/PAR of avs file?

Edit: I also did a test without AviSynth -- ffmpeg also couldn't handle the variable PAR with an avi file as a source.

[poisondeathray]

ffms2 doesn't show correct _SarNum/Den (probably older version, but new ones have problems with file names). LWLibavVideoSource does.

Yes, tehre were a few versions of ffms2 that had unicode support broken , not sure what the current status is

But ffmpeg with "-avisynth_flags all" can read DAR/PAR of avs file?

I didn't think it would, but it does for me at least for a single AR DV-AVI file


default

  Stream #0:0: Video: rawvideo (I420 / 0x30323449), yuv420p(bottom first), 720x576, 25 fps, 25 tbr, 25 tbn

-avisynth_flags all (as an input option)

  Stream #0:0: Video: rawvideo (I420 / 0x30323449), yuv420p(bottom first), 720x576, SAR 16:15 DAR 4:3, 25 fps, 25 tbr, 25 tbn

Can you verify if the switching AR is registered with lsmash , as least with the propshow()?

[rgr]

On every AVI file.



LWLibav works.

[jagabo]

LSMASHVideoSource only works for MP4 and MOV files.

[rgr]

Any ideas on how to properly convert such a file, preferably automatically?

I'm thinking about using CropResize -- I need to check if it can scale using frame PAR.
Or changing the resolution of the DAR 16:9 fragments to 960x576 and adding black bars to the DAR 4:3 fragments. And saving with PAR 16:15.
I'll have to think about it.

[dellsam34]

The best way is to separate the scenes, then add black pillars to all 4:3 scenes before they can be added back to their locations and joined with the rest of 16:9 scenes and processed into one single file.

At least that's what I would do if I have such tape. Use vdub to strip DV out, add black borders using the add border feature to the 4:3 frame one scene at a time and output into lossless AVI for each 4:3 segment, Then do the same thing for all 16:9 scenes, and rename all the output segments in a numerical order of their placement in the original footage, Use append feature in vdub to add all segments to the timeline and output as a single file and output in the same lossless AVI format, the entire process should be lossless.

From this large file you can de-interlace, resize to 1920x1080 and encode to your desired format.

[rgr]

I wonder if something like this has a chance of working. I did something like that, but it was changing one command depending on the height. From the Avisynth syntax I see that the chances are slim.

_prop("_SARNum") == 64 ? Eval("resize(960,576)" ? Eval("addBorders(left=120,right=120,color=0)")

Invented syntax -- that's how I imagine it.

[poisondeathray]

Here is a rough example using PropGetInt to parse the _SARNum, and it returns one version (b1) if SarNum is 64, otherwise (a1) . There is no error catching logic , and it's really rough

In this example, the "c" test clip is dark grey 4:3 + light grey 16:9 + dark grey 4:3 all at 720x576

Based on their SarNum value, those sections are replaced with a square pixel equivalent. Change the values as you see fit

blankclip(width=720,height=576,length=10,pixel_type="YV12",colors=[64,128,128]).killaudio()
propSet("_SARNum",16)  
propSet("_SARDen",15)  
#4:3
a=last

blankclip(width=720,height=576,length=10,pixel_type="YV12", colors=[192,128,128]).killaudio()
propSet("_SARNum",64)  
propSet("_SARDen",45)  
#16:9
b=last

a1=a.crop(8,0,-8,0).spline36resize(768,576).addborders(128,0,128,0).propSet("_SARNum",1).propSet("_SARDen",1)  
b1=b.crop(8,0,-8,0).spline36resize(1024,576).propSet("_SARNum",1).propSet("_SARDen",1)  

#testclip
c=a+b+a

scriptclip(c.spline36resize(1024,576), """propGetInt(c,"_SARNum") == 64? b1 : a1""")

[_Al_]

did not know that dv avi can have alternating aspect ratio, would be interesting if ffms2 or BestSource read those, in vapoursynth perhaps more likely (BestSource likely?):

from vapoursynth import core
from functools import partial
import havsfunc

#clip = core.ffms2.Source(r"E:\PAL_DV.avi")
clip = core.bs.VideoSource(r"E:\PAL_DV.avi")
clip = clip.std.SetFrameProps(_Matrix=5, _Primaries=5, _Transfer=5)
clip = clip.std.CropAbs(width=704, height=576, top=0, left=8)
# deinterlace or filters here
clip = havsfunc.QTGMC(clip, Preset='slow', TFF=False)
clip = clip.std.RemoveFrameProps(['_DurationNum', '_DurationDen'])

def compare_ar(n, f, clip):
    if int(f.props['_SARNum']) == 64:
        clip = clip.resize.Spline36(width=1024, height=576)
    else:
        clip = clip.resize.Spline36(width=768, height=576)
        clip = clip.std.AddBorders(left=128, right=128, color=(16,128,128))
    return clip

placeholder = clip.std.BlankClip(width=1024, height=576)
clip = core.std.FrameEval(placeholder, partial(compare_ar, clip=clip),  prop_src=clip)
clip = clip.std.SetFrameProps(_SARDen=1, _SARNum=1)
clip.set_output()

edit: I forgot to resize 4:3 parts

[rgr]

Thanks for the idea, I modified it a bit to suit my needs. I also replaced the filter with ConditionalFilter (according to the description it is more stable with multiple threads).

name="Tape 1 (01.09.2007 11.52).avi"

#4:3
a=LWlibavvideosource(name).QTGMC(preset="slow",TR2=3)#.propshow
a=AudioDub(a,LWlibavaudiosource(name))

#16:9
b=LWlibavvideosource(name).QTGMC(preset="slow",TR2=3)#.propshow

a1=a.addborders(120,0,120,0)
b1=b.spline36resize(960,576).propSet("_SARNum",16).propSet("_SARDen",15)

#scriptclip(b.spline36resize(960,576), """propGetInt(a,"_SARNum") == 16 ? a1 : b1""")
ConditionalFilter(a, a1, b1, """propGetInt(a,"_SARNum")""", "==", "16")

I just don't know why after adding audio the whole thing is unusable (either it works terribly slowly or it crashes VirtualDub)

Edit:
This is a problem with LWLibavAudioSource with uncompressed audio, I just remembered.

You need to use:

a=AudioDub(a,ffms2(name, atrack=-1))

* - I didn't use ffms2 for video because my older version doesn't read the properties of each frame and I couldn't find a new one that works properly with Unicode.

[rgr]

One more question for ffprobe experts. Is it possible to use this tool (or ffmpeg or avisynth) to check if a file has only one PAR or two?

Based on the following principle:
- all frames have the same PAR (16:15 or 64:45) - X value
- there are frames in the file with different PAR - Y value