Recherche avancée

Sur d’autres sites (8569)

• What is the correct way to calculate SSIM ?

  19 mai 2023, par jamacoe

  On Windows 10, I have installed ImageMagick, Version 7.1.1-Q16-HDRI.

  


  I have a high resolution .png image that I convert to jpg with 75% quality :

  


  magick myImage.png -quality 75 myImage.jpeg


  


  Now I want to calculate SSIM to judge the quality of the .jpeg in comparison to the original .png.

  


  First, I am using ffmpeg.exe version 4.2.3 from the very same ImageMagick installation :

  


  ffmpeg.exe -i myImage.jpeg -i myImage.png -lavfi ssim=stats_file=ssim_logfile.txt -f null -


  


  Results in ssim_logfile.txt :

  


  n:1 Y:0.959684 U:0.957603 V:0.976335 All:0.962112 (14.215018)


  


  So I assume that the SSIM value has been calculated as 0.962112

  


  Now I'll use magick to calculate the same metric :

  


  magick myImage.jpeg myImage.png -metric SSIM -compare -format "%[distortion]" info:


  


  Output is :

  


  0.925167


  


  That's a huge difference ! So I think I'm making an error here with one or the other way.

  


  What is the correct way to calculate SSIM ?

  


• How to find the offset by which the each video must be delayed to sync them perfectly ?

  19 janvier 2023, par PirateApp

  

  


  Let me explain my use case a bit here

  


  


  • We are 4 of us playing the same game

    


  


  • 3 of us recording mkv using OBS studio at 60 fps, 4th guy recording with some other tool at 30 fps

    


  


  • Each mission starts at a cutscene and ends with a cutscene

    


  


  • I would like to create a video like the image you see above starting at ending at the same points but the intermediate stuff is basically what each player is doing in the game

    


  


  • Currently, I follow a process slightly complicated to achieve this and was wondering if there is an easier way to do this

    


  


  • My current process

    


  


  • Take a screenshot from one of the videos of the cutscene

    


  


  


  Run a search for this screen inside the other videos using the command below

  


  ffmpeg 
  -i "video1.mkv"
  -r 1
  -loop 1
  -i 1.png
  -an -filter_complex "blend=difference:shortest=1,blackframe=90:32"
  -f null -


  


  


  • It gives me a result like this in each video

    


    [Parsed_blackframe_1 @ 0x600000c9c000] frame:263438 pblack:91 pts:4390633 t:4390.633000 type:P last_keyframe:263400

    


  


  


  Use the start time from each of the results to create a split screen video using the command below

  


  ffmpeg 
  -i first.mkv
  -i second.mkv
  -i third.mkv
  -i fourth.mp4
  -filter_complex " 
    nullsrc=size=640x360 [base];
    [0:v] trim=start=35.567,setpts=PTS-STARTPTS, scale=320x180 [upperleft]; 
    [1:v] trim=start=21.567,setpts=PTS-STARTPTS, scale=320x180 [upperright];
    [2:v] trim=start=41.233,setpts=PTS-STARTPTS, scale=320x180 [lowerleft]; 
    [3:v] trim=start=142.933333,setpts=PTS-STARTPTS, scale=320x180 [lowerright];
    [0:a] atrim=start=35.567,asetpts=PTS-STARTPTS [outa]; [base][upperleft] overlay=shortest=1 [tmp1];


  


  


  • As you can see, it is a complex process and depends completely a lot on what image I am capturing. Sometimes, I find out that stuff is still slightly off in the beginning or end because the images dont match a 100%. My guess is that the frame rate is different for each video not to mention 3 of them are mkv inputs and one is an mp4 input

    


  


  • Is there a better way to get the offset by how much each video should be moved to sync them perfectly ?

    


  


  • The only way that I can think of is to take 1 video

    


  


  • Take a starting timestamp and an ending timestamp, say with a total duration of 30s

    


  


  • Take the second video

    


  


  • Start from 0 to 30s and compare the frames in both videos, set a score

    


  


  • start from 0.001 to 30.001 and compare the frames, set a score

    


  


  • start from 0.002 to 30.002 and compare the frames, set a score

    


  


  • Basically increment the second video by 0.001 second each time and find out the part with the highest score

    


  


  • Any better way of doing this ? I need to run this on 100s if not 1000s of videos

    


  


  


• audio is not exact copy in ffmpeg proxies

  9 mars 2023, par 5Diraptor

  I always work with proxy footage in Premiere Pro, so I have a batch file set up to run any new raw footage through ffmpeg and produce a proxy version. I avoid creating proxies through Premiere Pro as it's slow, I have little control over the output, and often the proxies aren't compressed as much as they could be.

  


  This is the command I run to create proxies. It also overlays a watermark so that it is obvious when proxies are used.

  


  ffmpeg -i "original.MOV" -i "proxy_overlay.png" -filter_complex "overlay=0:0" -crf 20 -c:v libx264 -c:a copy "Proxy/proxy_for_original.mov"


  


  This has always been fine, but I've received some footage to work on from a different camera, and when I create proxies and try to attach them in Premiere Pro, I get the error : "Proxy Media and Full Resolution Media must have matching audio channels."

  


  

  


  When I compare the file properties in Premiere Pro, I see this :

  


  Original :

  


  

  


  Proxy :

  


  

  


  As you can see, the original contains 8x audio tracks and a timecode track, but the proxy version doesn't state this.

  


  I've been working on the ffmpeg command to try and retain the original audio details, but nothing I do seems to work. I thought that the command -c:a copy would copy the audio tracks across without any changes at all. However I can't seem to find a command that creates an exact copy of all audio tracks. Also does the timecode track matter, should I map this to the proxy as well ?