ffmpeg rtsp解码缓冲区太小 [英] Too small ffmpeg rtsp decoding buffer

问题描述

我正在使用ffmpeg在Android上解码rtsp，当图像快速或更高分辨率更新时，我很快看到像素化：

在谷歌搜索之后，我发现它可能与UDP缓冲区大小相关。然后，我重新编译ffmpeg库中的ffmpeg / libavformat / udp.c中的以下参数

  #define UDP_TX_BUF_SIZE 327680 
 #define UDP_MAX_PKT_SIZE 655360 
  

似乎有所改善，但在某些时候仍然会失败。任何想法哪个缓冲区应该增加，如何？

解决方案

对于我的问题（ http://libav-users.943685.n4.nabble.com/UDP- Stream-Read-Pixelation-Macroblock-Corruption-td4655270.html ），我试图从已经由其他人设置的多播UDP流中捕获。因为我没有能力混淆源码，所以我最终从使用libav切换到使用libvlc作为一个包装器，它的工作完美。以下是对我有用的总结：

stream.h：

  #include< vlc / vlc.h> 
 #include< vlc / libvlc.h> 
 
 struct ctx {
 uchar * frame; 
}; 
  

stream.cpp：

  void * lock（void * data，void ** p_pixels）{
 struct ctx * ctx =（struct ctx *）data; 
 * p_pixels = ctx-> frame; 
返回NULL; 
} 
 
 void unlock（void * data，void * id，void * const * p_pixels）{
 struct ctx * ctx =（struct ctx *）data; 
 uchar * pixels =（uchar *）* p_pixels; 
 assert（id == NULL）; 
} 
  

main.cpp：

  struct ctx * context =（struct ctx *）malloc（sizeof（* context））; 
 const char * const vlc_args [] = {-vvv，
-q，
--no-audio}; 
 libvlc_media_t * media = NULL; 
 libvlc_media_player_t * media_player = NULL; 
 libvlc_instance_t * instance = libvlc_new（sizeof（vlc_args）/ sizeof（vlc_args [0]），vlc_args）; 
 
 media = libvlc_media_new_location（实例，udp：//@123.123.123.123：1000）; 
 media_player = libvlc_media_player_new（instance）; 
 libvlc_media_player_set_media（media_player，media）; 
 libvlc_media_release（media）; 
 context-> frame = new uchar [height * width * 3]; 
 libvlc_video_set_callbacks（media_player，lock，unlock，NULL，context）; 
 libvlc_video_set_format（media_player，RV24，VIDEOWIDTH，VIDEOHEIGHT，VIDEOWIDTH * 3）; 
 libvlc_media_player_play（media_player）; 
  

I'm decoding rtsp on Android with ffmpeg, and I quickly see pixelization when the image updates quickly or with a high resolution:

After googling, I found that it might be correlated to the UDP buffer size. I have then recompiled the ffmpeg library with the following parameters inside ffmpeg/libavformat/udp.c

#define UDP_TX_BUF_SIZE 327680
#define UDP_MAX_PKT_SIZE 655360

It seems to improve but it still starts to fail at some point. Any idea which buffer I should increase and how?

解决方案

For my problem (http://libav-users.943685.n4.nabble.com/UDP-Stream-Read-Pixelation-Macroblock-Corruption-td4655270.html), I was trying to capture from a multicast UDP stream that had been set-up by someone else. Because I didn't have the ability to mess with the source, I ended up switching from using libav to using libvlc as a wrapper and it worked perfectly. Here is the summary of what worked for me:

stream.h:

#include <vlc/vlc.h>
#include <vlc/libvlc.h>

struct ctx {
   uchar* frame;
};

stream.cpp:

void* lock(void* data, void** p_pixels){
  struct ctx* ctx = (struct ctx*)data;
  *p_pixels = ctx->frame;
  return NULL;
}

void unlock(void* data, void* id, void* const* p_pixels){
  struct ctx* ctx = (struct ctx*)data;
  uchar* pixels = (uchar*)*p_pixels;
  assert(id == NULL);
}

main.cpp:

struct ctx* context = (struct ctx*)malloc(sizeof(*context));
const char* const vlc_args[] = {"-vvv",
                                 "-q",
                                 "--no-audio"};
libvlc_media_t* media = NULL;
libvlc_media_player_t* media_player = NULL;
libvlc_instance_t* instance = libvlc_new(sizeof(vlc_args) / sizeof(vlc_args[0]), vlc_args);

media = libvlc_media_new_location(instance, "udp://@123.123.123.123:1000");
media_player = libvlc_media_player_new(instance);
libvlc_media_player_set_media(media_player, media);
libvlc_media_release(media);
context->frame = new uchar[height * width * 3];
libvlc_video_set_callbacks(media_player, lock, unlock, NULL, context);
libvlc_video_set_format(media_player, "RV24", VIDEOWIDTH, VIDEOHEIGHT, VIDEOWIDTH * 3);
libvlc_media_player_play(media_player);

这篇关于ffmpeg rtsp解码缓冲区太小的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！