Recherche avancée

Sur d’autres sites (10263)

• FFMPEG zoompan image is cut off

  24 juillet 2018, par kklmarvelmedia

  Hi I’m using zoompan to resize an image overlay on a video using ffmpeg, but my image got cut off by the boundary after setting its size, like the following :

  ffmpeg -y -i 6.mp4 -loop 1 -i watermark.png -filter_complex "[1:v]zoompan=z=2:d=50:x=0:y=0:s='280x105'[1v];[0:v][1v]overlay=x=0:y=0" -t 8 test.mp4

  

  If I remove the size, it becomes really big covering all screen, like following :

  ffmpeg -y -i 6.mp4 -loop 1 -i watermark.png -filter_complex "[1:v]zoompan=z=2:d=50:x=0:y=0[1v];[0:v][1v]overlay=x=0:y=0" -t 8 test.mp4

  

  How do we zoom an overlay image while scaling its size ?

• How to stabilize inverse 360 video using stabilization algorithms ?

  11 août 2020, par Jimmy Leahy

  We have a 12-foot circular ring with about 50 cameras spaced evenly around the perimeter of the circle. These cameras are pointed at the exact center of the circle and are manually aligned. Our application snaps an image from all cameras simultaneously and makes a video from these images creating a matrix-like effect. I believe the effect is called Bullet Time.

  



  This process works well and does a great job of snapping a moment in time from all angles. The problem is that every camera is not perfectly aligned and the video ends up being slightly jerky due to the slightest differences in camera alignments.

  



  We need to stabilize the video. We are using FFmpeg's vidstabdetect and vidstabtransform to do this right now but the results are not desirable. From what I understand, these libraries are meant to be used with a video where the camera angle is static.

  



  Here are the exact commands :

  



  ffmpeg -i "input.mp4" -vf vidstabdetect=stepsize=32:shakiness=5:accuracy=15:result=transform_vectors.trf -f null - 
ffmpeg -i "input.mp4" -y -vf vidstabtransform=input=transform_vectors.trf:zoom=0:relative=1:interpol=bicubic:smoothing=10,unsharp=5:5:0.8:3:3:0.4 -b:v 100M -vcodec hevc_nvenc -tune film -an stabilized.mp4


  



  Are there any algorithms out there for these inverse or negative 260 videos ?

  


• Construct fictitious P-frames from just I-frames [closed]

  25 juillet 2024, par nilgirian

  Some context.. I saw this video recently https://youtu.be/zXTpASSd9xE?si=5alGvZ_e13w0Ahmb it's a continuous zoom into a fractal.

  


  I've been thinking a whole lot of how did they created this video 9 years ago ? The problem is that these frames are mathematically intensive to calculate back then and today still fairly really hard now.

  


  He states in the video it took him 33 hours to generate 1 keyframe.

  


  I was wondering how I would replicate that work. I know by brute force I can generate several images files (essentially each image would be an I-frame) and then ask ffmpeg to compress it into mp4 (where it will convert most of those images into P-frames). I know that. But if I did it that way I calculated it'd take me 6.5 years to render that 9min video (at 30fps, 9 years ago).

  


  So I imagine he only generated I-frames to cut down on time. And then this person somehow created fictitious P-frames in-between. Given that frame-to-frame are similar this seems like it should be doable since you're just zooming in. If he only generated just the I-frames at every 1 second (at 30fps) that work could be cut down to just 82 days.

  


  So if I only want to generate the images that will be used as I-frames could ffmpeg or some other program just automatically make a best guess to generate fictitious P-frames for me ?