Recherche avancée

Sur d’autres sites (5160)

• ffmpeg transcoding from mkv to webm using av1_vaapi [closed]

  7 janvier, par user248979

  I'm trying to convert HEVC encoded mkv videos into webm files using GPU acceleration.
So far I've tried

  


  ffmpeg -hwaccel vaapi -hwaccel_device /dev/dri/renderD128 -hwaccel_output_format vaapi -i "input.mkv" -c:v av1_vaapi output.webm

  


  the result audio works but the video is completely garbled. I've tried a different way :

  


  ffmpeg -vaapi_device /dev/dri/renderD128 -i input.mkv -vf 'format=nv12,hwupload' -c:v av1_vaapi output.webm

  


  But the result is the same. What am I doing wrong ?

  


  vainfo

  


  vainfo: VA-API version: 1.22 (libva 2.22.0)
vainfo: Driver version: Mesa Gallium driver 24.2.8-1 for AMD Radeon RX 7900 XT (radeonsi, navi31, LLVM 19.1.4, DRM 3.59, 6.12.6-amd64)
vainfo: Supported profile and entrypoints
      VAProfileH264ConstrainedBaseline: VAEntrypointVLD
      VAProfileH264ConstrainedBaseline: VAEntrypointEncSlice
      VAProfileH264Main               : VAEntrypointVLD
      VAProfileH264Main               : VAEntrypointEncSlice
      VAProfileH264High               : VAEntrypointVLD
      VAProfileH264High               : VAEntrypointEncSlice
      VAProfileHEVCMain               : VAEntrypointVLD
      VAProfileHEVCMain               : VAEntrypointEncSlice
      VAProfileHEVCMain10             : VAEntrypointVLD
      VAProfileHEVCMain10             : VAEntrypointEncSlice
      VAProfileJPEGBaseline           : VAEntrypointVLD
      VAProfileVP9Profile0            : VAEntrypointVLD
      VAProfileVP9Profile2            : VAEntrypointVLD
      VAProfileAV1Profile0            : VAEntrypointVLD
      VAProfileAV1Profile0            : VAEntrypointEncSlice
      VAProfileNone                   : VAEntrypointVideoProc


  


  OS:Debian-testing

  


• vulkan_decode : remove informative queries

  24 décembre 2024, par Lynne

  vulkan_decode : remove informative queries
  We queried the decoder whether it was able to decode sucessfully, but
  
  since we operated asynchronously, we weren't able to do anything with
  
  this information but let the user know decoding failed for the previous
  
  frame(s).
  Since we parse the slice headers ourselves and we're reasonably sure we
  
  can decode before actually starting to decode, this was rarely triggered
  
  on corrupt data, and hardware's understanding of whether there was an error
  
  or not is vague.
  There's also a semantic problem with our use of the queries - if there's
  
  a seek, we flush, but what happens to the queries is vague according to
  
  the spec. Most hardware dealt fine, since queries are nothing more than
  
  GPU memory with integers stored. But with Intel, they seem to be more of
  
  a register to which a driver must keep track of, leading to issues if there's
  
  been a reset (seek) and we query the previous submission before the seek.
  Just get rid of them. The query code is still used in encoding.
  This fixes seeking with HEVC and AV1 on Intel.
  

  • [DH] libavcodec/vulkan_decode.c

• Record video stream in rust

  19 novembre 2024, par El_Loco

  I have bought a stereo camera with global shutter and a frame rate of at most 120 fps. https://www.amazon.com/dp/B0D8T3ZSL4?ref_=pe_386300_442618370_TE_sc_as_ri_0#

  


  My next step is to write a program that can show and record a video with desired fps and resolution.

  


  use opencv::{&#xA;    core, highgui,&#xA;    prelude::*,&#xA;    videoio::{self, VideoCapture},&#xA;    Result,&#xA;};&#xA;&#xA;fn open_camera() -> Result<videocapture> {&#xA;    let capture = videoio::VideoCapture::new(2, videoio::CAP_ANY)?;&#xA;    return Ok(capture);&#xA;}&#xA;fn main() -> Result&lt;()> {&#xA;    let window = "video capture";&#xA;    highgui::named_window(window, highgui::WINDOW_AUTOSIZE)?;&#xA;    let mut cam = open_camera()?;&#xA;    let opened = videoio::VideoCapture::is_opened(&amp;cam)?;&#xA;    if !opened {&#xA;        panic!("Unable to open default camera!");&#xA;    }&#xA;    let width = 3200.0;&#xA;    let height = 1200.0;&#xA;    cam.set(videoio::CAP_PROP_FRAME_WIDTH, width)?;&#xA;    cam.set(videoio::CAP_PROP_FRAME_HEIGHT, height)?;&#xA;&#xA;    // Set the frame rate (FPS)&#xA;    let fps = 60.0;&#xA;        &#xA;    let fourcc = videoio::VideoWriter::fourcc(&#x27;M&#x27;, &#x27;J&#x27;, &#x27;P&#x27;, &#x27;G&#x27;)?;&#xA;    let mut writer = videoio::VideoWriter::new(&#xA;        "video_output.avi",&#xA;        fourcc,&#xA;        fps,&#xA;        core::Size::new(width as i32, height as i32),&#xA;        true,&#xA;    )?;&#xA;&#xA;    if !writer.is_opened()? {&#xA;        println!("Error: Could not open the video writer.");&#xA;    }&#xA;&#xA;    let mut frame = core::Mat::default();&#xA;    let mut ctr = 0;&#xA;    while cam.read(&amp;mut frame)? {&#xA;        if frame.empty() {&#xA;            break;&#xA;        }&#xA;        writer.write(&amp;frame)?;&#xA;        highgui::imshow(window, &amp;frame)?;&#xA;        &#xA;        let key = highgui::wait_key(1)?;&#xA;        if key > 0 {&#xA;            break;&#xA;        }&#xA;        ctr &#x2B;= 1;&#xA;        if ctr == 600 {&#xA;            break;&#xA;        }&#xA;    }&#xA;    cam.release()?;&#xA;    writer.release()?;&#xA;    Ok(())&#xA;}&#xA;</videocapture>

  


  When I run this code the frame rate is terrible. Like 1 fps or something. For debugging I tried to run in cheese. There I got 30 fps with full resolution 3200x1200. But I cannot change the fps to 60 fps what I can see.

  


  Then I tried to capture a video using ffmpeg :

  


  ffmpeg -f v4l2 -framerate 60 -video_size 3200x1200 -i /dev/video2 output.mp4

  


  With the following output :

  


  [video4linux2,v4l2 @ 0x5a72cbbd1400] The driver changed the time per frame from 1/60 to 1/2
Input #0, video4linux2,v4l2, from '/dev/video2':
  Duration: N/A, start: 2744.250608, bitrate: 122880 kb/s
  Stream #0:0: Video: rawvideo (YUY2 / 0x32595559), yuyv422, 3200x1200, 122880 kb/s, 2 fps, 2 tbr, 1000k tbn
File 'output.mp4' already exists. Overwrite? [y/N]


  


  The frame rate is lowered to 2 fps.

  


  Then I tried to run v4l2-ctl --list-formats-ext -d 2 with the following output :

  


  ioctl: VIDIOC_ENUM_FMT
        Type: Video Capture

        [0]: 'MJPG' (Motion-JPEG, compressed)
                Size: Discrete 3200x1200
                        Interval: Discrete 0.017s (60.000 fps)
                        Interval: Discrete 0.033s (30.000 fps)
                        Interval: Discrete 0.040s (25.000 fps)
                        Interval: Discrete 0.050s (20.000 fps)
                        Interval: Discrete 0.067s (15.000 fps)
                        Interval: Discrete 0.100s (10.000 fps)
                Size: Discrete 2560x720
                        Interval: Discrete 0.017s (60.000 fps)
                        Interval: Discrete 0.033s (30.000 fps)
                        Interval: Discrete 0.040s (25.000 fps)
                        Interval: Discrete 0.050s (20.000 fps)
                        Interval: Discrete 0.067s (15.000 fps)
                        Interval: Discrete 0.100s (10.000 fps)
                Size: Discrete 1600x600
                        Interval: Discrete 0.008s (120.000 fps)
                        Interval: Discrete 0.017s (60.000 fps)
                        Interval: Discrete 0.033s (30.000 fps)
                        Interval: Discrete 0.040s (25.000 fps)
                        Interval: Discrete 0.050s (20.000 fps)
                        Interval: Discrete 0.067s (15.000 fps)


  


  I then tried to open the camera using qv4land there it seemed to work. Does not seem like I can record a video though.

  


  I am using Rust to learn. I want to be able to programmatically be able to record a video somehow and then do computer vision. The easiest would be to do it in Rust. But other solutions are ok.

  


  Edit
I have found some more this morning :

  


  v4l2-ctl -d 2 --list-formats-ext
ioctl: VIDIOC_ENUM_FMT
    Type: Video Capture

    [0]: 'MJPG' (Motion-JPEG, compressed)
        Size: Discrete 3200x1200
            Interval: Discrete 0.017s (60.000 fps)
            Interval: Discrete 0.033s (30.000 fps)
            Interval: Discrete 0.040s (25.000 fps)
            Interval: Discrete 0.050s (20.000 fps)
            Interval: Discrete 0.067s (15.000 fps)
            Interval: Discrete 0.100s (10.000 fps)

    [1]: 'YUYV' (YUYV 4:2:2)
        Size: Discrete 3200x1200
            Interval: Discrete 0.500s (2.000 fps)
        Size: Discrete 2560x720
            Interval: Discrete 0.500s (2.000 fps)
        Size: Discrete 1600x600
            Interval: Discrete 0.100s (10.000 fps)


  


  I also found here that order of flags was important for ffmpeg. Running this I can actually record a video with 60 fps :

  


  ffmpeg -framerate 60 -f v4l2 -video_size 3200x1200 -input_format mjpeg  -i /dev/video2 output.avi

  


  A drawback is that the images does not look very sharp. You can clearly see the pixels. (I am new to video formats etc as well. Before it has just worked.)

  


  If I change from avito mkvit is slow again.

  


  In the link above I also saw a suggestion to first do :

  


  ffmpeg -framerate 60 -f v4l2 -video_size 3200x1200 -input_format mjpeg  -i /dev/video2 -c copy mjpeg.mkv

  


  and then :

  


  ffmpeg -i mjpeg.mkv -c:v libx264 -crf 23 -preset medium -pix_fmt yuv420p out.mkv

  


  which worked. But I am not sure those flags are ideal for the camera I have. I think it is a good start to make it run as expected using command line and ffmpeg. So I know what format to use and that it actually works as intended before doing it programmatically.