Recherche avancée

Sur d’autres sites (9628)

• Model the loss of video codec

  10 janvier 2024, par Monaco

  I want to use a neural network to model the error loss after video encoding. The modeling process is as follows :

I use ffmpeg to encode and decode video frames. Since this process is not implemented using tensors in PyTorch and cannot compute gradients, I have to separately implement a neural network in PyTorch to enable gradient backpropagation. However, it turns out that the neural network cannot effectively learn the video encoding.

  


  I want to know if there are currently any implementations of video encoders or decoders that support backpropagation of gradients. I don't necessarily need to update the parameters of the encoder/decoder, but I want it to support gradient backpropagation so that I can use it for various tasks related to deep learning.

  


• Low FPS output using ffmpeg and a raspi camera

  23 novembre 2019, par Newe

  I am building a surveillance camera for a school project, which is based on a raspberry pi and a infrared raspberry Pi camera.

  I am capturing the camera’s video stream and outputting it as an HLS stream directly from ffmpeg. However, the resulting video is really low fps ( 5 at max)

  Strangely, raspivid manages to ouput a 60fps 720p stream without any problem, but when put through ffmpeg for streaming, the video is cropped in half and i cannot get it to show up entirely.

  Here is the ffmpeg command i use :

  #!/bin/bash
  
  ffmpeg -v verbose \
  
  -re \
  
  -i /dev/video0 \
  
  -c:v libx264 \
  
  -an \
  
  -f hls \
  
  -g 10  \
  
  -sc_threshold 0 \
  
  -hls_time 1 \
  
  -hls_list_size 4 \
  
  -hls_delete_threshold 1 \
  
  -hls_flags delete_segments \
  
  -hls_start_number_source datetime \
  
  -hls_wrap 15 \
  
  -preset superfast \
  
  -start_number 1 \
  
  /home/pi/serv/assets/stream.m3u8

  And the resulting log output (notice the fps)

  

  Here is the command using raspivid that i tested, based on a blog post i read :

  raspivid -n \
  
  -t 0 \
  
  -w 960 \
  
  -h 540 \
  
  -fps 25 \
  
  -o - | ffmpeg \
  
  -v verbose \
  
  -i - \
  
  -vcodec copy \
  
  -an \
  
  -f hls \
  
  -g 10 \
  
  -sc_threshold 0 \
  
  -hls_time 1 \
  
  -hls_list_size 4 \
  
  -hls_delete_threshold 1 \
  
  -hls_flags delete_segments \
  
  -hls_start_number_source datetime \
  
  -hls_wrap 15 \
  
  -preset superfast \
  
  -start_number 1 \
  
  /home/pi/serv/assets/stream.m3u8

  I am not a ffmpeg expert and am open to any suggestions that would help improve the stream’s quality and stability :)

• Converting mp4 to multiple resolutions using FFMPEG

  2 avril 2018, par Edwin Flataunet

  I am developing a E-Learning website where people can upload videos.

  When someone uploads a video, they usually upload it in 1080p full HD which is good.
  Now the issue is that when people watch the videos, they are only in 1080p and some people have bad internet (including myself) so watching a video in 1080p is not optimal.

  So I assume that converting the video to different formats (720p, 360 etc..) is the best way to go here.
  So I tried using FFMpeg, and it works, but its really slow, especially since some videos are over 10 minutes long and over 1gb in size.

  I use this command in FFMPEG :

  ffmpeg -i video.mp4 -vf scale:1280:720 -strict -2 output.mp4

  This works, but its really slow.

  Is there any better way to do this ? Since some people upload 5-10 videos and every video has to be in 3 different formats (1080p(original) 720p, 360p).

  Can someone give me some guidelines how to tackle this issue, as this kind of stops the website from progressing atm.

  Thanks