Sur d’autres sites (9323)

• Simply beyond ridiculous

  7 mai 2010, par Dark Shikari — H.265, speed

  For the past few years, various improvements on H.264 have been periodically proposed, ranging from larger transforms to better intra prediction. These finally came together in the JCT-VC meeting this past April, where over two dozen proposals were made for a next-generation video coding standard. Of course, all of these were in very rough-draft form ; it will likely take years to filter it down into a usable standard. In the process, they’ll pick the most useful features (hopefully) from each proposal and combine them into something a bit more sane. But, of course, it all has to start somewhere.

  A number of features were common : larger block sizes, larger transform sizes, fancier interpolation filters, improved intra prediction schemes, improved motion vector prediction, increased internal bit depth, new entropy coding schemes, and so forth. A lot of these are potentially quite promising and resolve a lot of complaints I’ve had about H.264, so I decided to try out the proposal that appeared the most interesting : the Samsung+BBC proposal (A124), which claims compression improvements of around 40%.

  The proposal combines a bouillabaisse of new features, ranging from a 12-tap interpolation filter to 12thpel motion compensation and transforms as large as 64×64. Overall, I would say it’s a good proposal and I don’t doubt their results given the sheer volume of useful features they’ve dumped into it. I was a bit worried about complexity, however, as 12-tap interpolation filters don’t exactly scream “fast”.

  I prepared myself for the slowness of an unoptimized encoder implementation, compiled their tool, and started a test encode with their recommended settings.

  I waited. The first frame, an I-frame, completed.

  I took a nap.

  I waited. The second frame, a P-frame, was done.

  I played a game of Settlers.

  I waited. The third frame, a B-frame, was done.

  I worked on a term paper.

  I waited. The fourth frame, a B-frame, was done.

  After a full 6 hours, 8 frames had encoded. Yes, at this rate, it would take a full two weeks to encode 10 seconds of HD video. On a Core i7. This is not merely slow ; this is over 1000 times slower than x264 on “placebo” mode. This is so slow that it is not merely impractical ; it is impossible to even test. This encoder is apparently designed for some sort of hypothetical future computer from space. And word from other developers is that the Intel proposal is even slower.

  This has led me to suspect that there is a great deal of cheating going on in the H.265 proposals. The goal of the proposals, of course, is to pick the best feature set for the next generation video compression standard. But there is an extra motivation : organizations whose features get accepted get patents on the resulting standard, and thus income. With such large sums of money in the picture, dishonesty becomes all the more profitable.

  There is a set of rules, of course, to limit how the proposals can optimize their encoders. If different encoders use different optimization techniques, the results will no longer be comparable — remember, they are trying to compare compression features, not methods of optimizing encoder-side decisions. Thus all encoders are required to use a constant quantizer, specified frame types, and so forth. But there are no limits on how slow an encoder can be or what algorithms it can use.

  It would be one thing if the proposed encoder was a mere 10 times slower than the current reference ; that would be reasonable, given the low level of optimization and higher complexity of the new standard. But this is beyond ridiculous. With the prize given to whoever can eke out the most PSNR at a given quantizer at the lowest bitrate (with no limits on speed), we’re just going to get an arms race of slow encoders, with every company trying to use the most ridiculous optimizations possible, even if they involve encoding the frame 100,000 times over to choose the optimal parameters. And the end result will be as I encountered here : encoders so slow that they are simply impossible to even test.

  Such an arms race certainly does little good in optimizing for reality where we don’t have 30 years to encode an HD movie : a feature that gives great compression improvements is useless if it’s impossible to optimize for in a reasonable amount of time. Certainly once the standard is finalized practical encoders will be written — but it makes no sense to optimize the standard for a use-case that doesn’t exist. And even attempting to “optimize” anything is difficult when encoding a few seconds of video takes weeks.

  Update : The people involved have contacted me and insist that there was in fact no cheating going on. This is probably correct ; the problem appears to be that the rules that were set out were simply not strict enough, making many changes that I would intuitively consider “cheating” to be perfectly allowed, and thus everyone can do it.

  I would like to apologize if I implied that the results weren’t valid ; they are — the Samsung-BBC proposal is definitely one of the best, which is why I picked it to test with. It’s just that I think any situation in which it’s impossible to test your own software is unreasonable, and thus the entire situation is an inherently broken one, given the lax rules, slow baseline encoder, and no restrictions on compute time.

• How to get video pixel location from screen pixel location ?

  22 février 2024, par AmLearning

  Wall of Text so I tried breaking it up into sections to make it better sorry in advance

  


  The problem

  


  I have some video files that I am reading with ffmpeg to get the colors at specific pixels, and all seems well, but I just ran into a problem with finding the right pixel to input. I realized (or mistakingly believe) that the pixel location (x,y) on the screen will be different than the local pixel location so to speak of the video (ie. If I want to get pixel 50,0 of the video that will be different than my screen's pixel 50,0 because the resolutions don't match). I was trying to think of a way to convert my screen's pixel location into the "local pixel location", and I have two ideas but I am not sure if any of them is any good. Note I am currently using cmd+shift+4 on macos to get the screen coordinates and the video is playing fullscreen like in the screenshot below.

  


  Ideas

  


  


  1. If I manually measure and account for this vertical offset, would it effectively convert the screen coordinate into the "local" one ?

     


  


  2. If I instead adjust my SwsContext to put the destination height and width as that of my screen, will it effectively replace the need to convert screen coordinates to the video coordinates ?

     


  


  


  Problems with the Ideas

  


  The problems I see with the first solution are that I am assuming there is no hidden horizontal offset (or conversely that all of the width of the video is actually renderable on the screen). Additionally, this solution would only get an approximate result as I would need to manually measure the offsets, screen width, and screen height using the method I currently am using to get the screen coordinates.

  


  With the second solution, aside from the question of if it will even work, the problem becomes that I can no longer measure what the screen coordinates I want are because I can't seem to get rid of those black bars in VLC.

  


  Some Testing I did

  


  Given that if the black bars are part of the video itself, my entire problem would be fixed (maybe ?) I tried seeing if the black bars were part of the video, and when I looked at the frame data's first pixel, it was black. The problem then is that if the black bars are entirely part of the video, then why are the colors I get for some pixels slightly off (I am checking with ColorSync Utility). These colors aren't just slightly off as in wrong but it seems more that they belong to a slightly offset region of the video.

  


  However, this may be somewhat explained if ffmpeg reads right to left. When I put the top left corner of the video into the program and looked again at the pixel data in the frame for that location (location again was calculated by assuming the video location would be the same as the screen location) instead of getting white, I got a bluish color much like the glove in the top right corner.

  


  The Watered Down Code

  


      struct SwsContext *rescaler = NULL;
    rescaler = sws_getContext(codec_context->width, codec_context->height, codec_context->pix_fmt, codec_context->width, codec_context->height, AV_PIX_FMT_RGB0, SWS_FAST_BILINEAR, NULL, NULL, 0);

// Get Packets (containers for frames but not guaranteed to have a full frame) and Frames
    while (av_read_frame(avformatcontext, packet) >= 0)
    {
        
        // determine if packet is video packet
        if (packet->stream_index != video_index)
        {
            continue;
        }
        
        // send packet to decoder
        if (avcodec_send_packet(codec_context, packet) < 0)
        {
            perror("Failed to decode packet");
        }
        
        // get frame from decoder
        int response = avcodec_receive_frame(codec_context, frame);
        if (response == AVERROR(EAGAIN))
        {
            continue;
        }
        else if (response < 0)
        {
            perror("Failed to get frame");
        }
        
        // convert frame to RGB0 colorspace 4 bytes per pixel 1 per channel
        response = sws_scale_frame(rescaler, scaled_frame, frame);
        if(response < 0){
            perror("Failed to change colorspace");
        }
        // get data and write it
        int pixel_number = y*(scaled_frame->linesize[0]/4)+x; // divide by four gets pixel linesize (4 byte per pixel)
        int byte_number = 4*(pixel_number-1); // position of pixel in array
        // start of debugging things
        int temp = scaled_frame->data[0][byte_number]; // R
        int one_after = scaled_frame->data[0][byte_number+1]; // G
        int two_after = scaled_frame->data[0][byte_number+2]; // B
        int als; // where i put the breakpoint
        // end of debugging things
    }


  


  In Summary

  


  I have no idea what is happening.

  


  I take the data for a pixel and compare it to what colorsync utility says should be there, but it is always slightly off as though the pixel I was actually reading was offset from what I thought I was reading. Therefore, I want to find a way to get the pixel location in a video given a screen coordinate when the video is in fullscreen, but I have no idea how to (aside from a few ideas that are probably bad at best).

  


  Also does FFMPEG put the frame data right to left ?

  


  A Video Better Showing My Problem

  


  https://www.youtube.com/watch?v=NSEErs2lC3A

  


• Best ffmpeg options for creating TikTok-like app ? [closed]

  27 août 2022, par nickcoding

  My app is essentially vertical videos, so I'd like the videos to more or less be cropped to that dimension (especially in the case they are 'landscape mode' videos. In the below picture, if the video is that horizontal black rectangle, we should just crop the video to that dimension of the phone (the red vertical rectangle). I've tried messing around with aspect ratios, but can't seem to get it to work right (where the video is essentially cropped to the correct aspect ratio), but not necessarily hard-coding the dimensions (this would look bad on different size devices).

  


  I'm using this module : https://www.npmjs.com/package/fluent-ffmpeg

  


  How should I do this ?

  


  https://i.stack.imgur.com/bb8qR.png

  


  var ffmpeg = require('fluent-ffmpeg');

  ffmpeg('./original.mp4')
    .fps(25)
    .videoCodec('libx264')
    // here, I've tried .withAspectRatio('16:9') and playing around with .size(), but that seems to do nothing.
    .save('./new.mp4');