Reputation: 603
I have a source code in C++ using libavcodec for decoding h264 rtsp stream frames. This source code was written using ffmpeg 1.1. Now when I upgraded to ffmpeg 3.3, all things seem to work correctly except that decoding frames not work. In old version, I was using avcodec_decode_video2. After upgrading, avcodec_decode_video2 always sets got_picture to 0 and return value is equal to the size of the input packet (which means all data is used). And never a frame is decoded. I have also removed avcodec_decode_video2 and done decoding with avcodec_send_packet and avcodec_receive_frame, but avcodec_send_packet always returns 0 and avcodec_receive_frame always returns -11 (EAGAIN).
This is the code I use for decoding:
#include "stdafx.h"
#include <stdlib.h>
#include <string>
#include <iostream>
using namespace std;
extern "C"{
#include "libavformat/avformat.h"
#include "libavcodec/avcodec.h"
#include "libswscale/swscale.h"
#include "libavutil/pixfmt.h"
}
int extraDataSize;
static const int MaxExtraDataSize = 1024;
uint8_t extraDataBuffer[MaxExtraDataSize];
void AddExtraData(uint8_t* data, int size)
{
auto newSize = extraDataSize + size;
if (newSize > MaxExtraDataSize){
throw "extradata exceeds size limit";
}
memcpy(extraDataBuffer + extraDataSize, data, size);
extraDataSize = newSize;
}
int _tmain(int argc, _TCHAR* argv[])
{
std::string strFramesPath("g:\\frames\\");
AVCodec* avCodec;
AVCodecContext* avCodecContext;
AVFrame* avFrame;
AVCodecID codecId = AV_CODEC_ID_H264;
unsigned char sprops_part_1[9] = { 0x27, 0x42, 0x80, 0x1f, 0xda, 0x02, 0xd0, 0x49, 0x10 };
unsigned char sprops_part_2[4] = { 0x28, 0xce, 0x3c, 0x80 };
av_register_all();
avcodec_register_all();
avCodec = avcodec_find_decoder(codecId);
avCodecContext = avcodec_alloc_context3(avCodec);
if (!avCodecContext)
{
cout << "avcodec_alloc_context3 failed." << endl;
return 0;
}
uint8_t startCode[] = { 0x00, 0x00, 0x01 };
// sprops
{
// sprops 1
AddExtraData(startCode, sizeof(startCode));
AddExtraData(sprops_part_1, 9);
// sprops 2
AddExtraData(startCode, sizeof(startCode));
AddExtraData(sprops_part_2, 4);
avCodecContext->extradata = extraDataBuffer;
avCodecContext->extradata_size = extraDataSize;
}
AddExtraData(startCode, sizeof(startCode));
avCodecContext->flags = 0;
if (avcodec_open2(avCodecContext, avCodec, NULL) < 0)
{
cout << "failed to open codec" << endl;
return 0;
}
avFrame = av_frame_alloc();
if (!avFrame)
{
cout << "failed to alloc frame" << endl;
return 0;
}
void *buffer = malloc(100 * 1024); // 100 KB buffer - all frames fit in this buffer
for (int nFrameIndex = 0; nFrameIndex < 257; nFrameIndex++)
{
std::string strFilename = std::string("g:\\frames\\" + std::to_string(nFrameIndex));
FILE* f = fopen(strFilename.c_str(), "rb");
fseek(f, 0, SEEK_END);
long nFileSize = ftell(f);
fseek(f, 0, SEEK_SET);
size_t nReadSize = fread(buffer, 1, nFileSize, f);
// cout << strFilename << endl;
if (nReadSize != nFileSize)
{
cout << "Error reading file data" << endl;
continue;
}
AVPacket avpkt;
avpkt.data = (uint8_t*)buffer;
avpkt.size = nReadSize;
while (avpkt.size > 0)
{
int got_frame = 0;
auto len = avcodec_decode_video2(avCodecContext, avFrame, &got_frame, &avpkt);
if (len < 0) {
//TODO: log error
cout << "Error decoding - error code: " << len << endl;
break;
}
if (got_frame)
{
cout << "* Got 1 Decoded Frame" << endl;
}
avpkt.size -= len;
avpkt.data += len;
}
}
getchar();
return 0;
}
Test frames data can be downloaded from this link: Frames.zip (~3.7MB)
I have used windows builds from Builds - Zeranoe FFmpeg If you copy paste this code into your IDE, the code compiles successfully. Using libavcodec new versions, no frame is decoded. Using old version of libavcodec (20141216-git-92a596f), decoding starts when feed frame 2.
Any ideas?
Upvotes: 1
Views: 917
Reputation: 5334
FFmpeg/libAV description is quite complex and probably not the best one in terms of understandability (if not to say it's bad). You should read it carefully.
However, what you are missing is to init the AVPacket
by using av_init_packet
. Or use av_packet_alloc()
to create the packet which will initialize the packet (and is better for duplicating it later for encoding).
// code ...
AVPacket avpkt;
av_init_packet(&avpkt); // initialize packet
avpkt.data = (uint8_t*)buffer;
avpkt.size = nReadSize;
// code ...
or
// code ...
AVPacket* avpkt = av_packet_alloc();
avpkt->data = (uint8_t*)buffer;
avpkt->size = nReadSize;
// use ...
auto len = avcodec_decode_video2(avCodecContext, avFrame, &got_frame, avpkt);
Using av_packet_alloc()
is also more consistent within the code since you also use AVFrame
as a pointer.
Remember to free the packets/frames/pictures as well, otherwise you will have memory leaks.
Also here is a (more or less) good/"up-to-date" tutorial for ffmpeg.
Upvotes: 1