Recherche avancée

Sur d’autres sites (6407)

• How to extract a fixed set of frames from a live video stream for machine learning prediction in PyTorch ?

  12 avril 2022, par Samay Lakhani

  I recently created a Video Swin Transformer model that takes in a ([batch_size], 3, 32, 224, 224) [batch_size, channel, temporal_dim, height, width] tensor for video and outputs logits. The goal is to have the model predict on a live stream from a camera. Is there any way to capture the fixed sequence of 32 frames repetitively and have the model predict on a live stream. If prediction time is longer than 32 frames, can I stretch out the frames over a longer time period like a minute ? Thanks.

  


• Working example on live streaming using FFMPEG+MP4Box+Dash.js

  27 février 2017, par BartMao

  I’m stucked on this issue for a couple of days, couldn’t find a perfect working example.

  Here’s my current method :

  Step 1.

  ffmpeg -re -i sample.mp4 -threads 0 -y \
  
  -map 0:1 -vn -acodec aac -strict -2 -ar 48000 -ac 2 \
  
  -f segment -segment_time 4 -segment_format mpegts Dash\audio_%d.ts \
  
  -map 0:0 -vcodec libx264 -vprofile baseline -preset veryfast \
  
  -s 640x360 -vb 512k -bufsize 1024k -maxrate 512k -level 31 -keyint_min 25 -g 25 \
  
  -sc_threshold 0 -an -bsf h264_mp4toannexb -flags -global_header \
  
  -f segment -segment_time 4 -segment_format mpegts Dash\video_%d.ts

  Step 2.

  mp4box -add Dash\audio_%d.ts Output\audio_%d.ts.mp4
  
  mp4box -add Dash\video_%d.ts Output\video_%d.ts.mp4

  Step 3.

  mp4box -dash-ctx Output\dash-live.txt -dash 4000 \
  
  -rap -ast-offset 12 -no-frags-default -bs-switching no \
  
  -min-buffer 4000 -url-template -time-shift 1800 \
  
  -segment-name live_$RepresentationID$_ -out Output\live \
  
  -dynamic -mpd-refresh 2 -subsegs-per-sidx -1 Output\audio_%d.ts.mp4:id=audio Output\video_%d.ts.mp4:id=video

  I played the live.mpd generated using Dash.js, the player paused a long time before actually playing, and freeze every time getting a new m4s file.

  Are there any wrong on my commands ? Or it would be great if can provide any other good examples.

• Http Live Streaming - Segmenting mp3 on Linux

  14 mai 2012, par krisbulman

  I simply want to segment an mp3 for HTTP Live Streaming in any linux distro (preferably CentOS) for the purpose of audio streaming to an iOS app.

  Out of the linux segmenters, I can get the following to compile in CentOS.

  1. http://wiki.andy-chu.com/doku.php?id=http_live_streaming (not sure last time this was updated)

  2. m3u8-segmenter on github (updated months ago)

  3. https://github.com/carsonmcdonald/HTTP-Live-Video-Stream-Segmenter-and-Distributor [ruby wrappers + c] (last updated 2 years ago, and a v2 branch 9 months old)

  In order to prep the file for segmenting, here is the ffmpeg conversion string to generate a valid ts file :

  $ ffmpeg -er 4 -i input.mp3 -f mpegts -acodec libmp3lame -ar 22050 -ab 32k -vn output.ts

  Each of the segmenters require various input switches, all quite simple, and all crash out with a seg fault. #2 actually does some segmenting, but faults after 56 segments every time. I've tried various mp3s with the same results. The issue queues for 2 & 3 are full, with no responses in months of the same issues.

  Others must be doing this in a live production environment that isn't running OSX.. what are your methods ?