I am using FFmpeg's C API to push video streams rtmp://.... into an SRS server.
The input stream is an MP4 file named juren-30s.mp4.
The output stream is also an MP4 file named juren-30s-5.mp4.
My piece of code (see further down) works fine when used in the following steps:
mp4 -> demux -> decode -> rgb images -> encode -> mux -> mp4.
Problem:
When I changed the output stream to an online RTMP url named rtmp://ip:port/live/stream_nb_23 (just an example, you can change it according to your server and rules.)
result: This code would be corrupted mp4 -> rtmp(flv).
What I've tried:
Changing the output format
I changed the output format param to become flv when I initialized the avformat_alloc_output_context2. But this didn't help.
Debugging the output
When I executed ffprobe rtmp://ip:port/live/xxxxxxx, I got the following errors and did not know why:
[h264 @ 0x55a925e3ba80] luma_log2_weight_denom 12 is out of range
[h264 @ 0x55a925e3ba80] Missing reference picture, default is 2
[h264 @ 0x55a925e3ba80] concealing 8003 DC, 8003 AC, 8003 MV errors in P frame
[h264 @ 0x55a925e3ba80] QP 4294966938 out of range
[h264 @ 0x55a925e3ba80] decode_slice_header error
[h264 @ 0x55a925e3ba80] no frame!
[h264 @ 0x55a925e3ba80] luma_log2_weight_denom 21 is out of range
[h264 @ 0x55a925e3ba80] luma_log2_weight_denom 10 is out of range
[h264 @ 0x55a925e3ba80] chroma_log2_weight_denom 12 is out of range
[h264 @ 0x55a925e3ba80] Missing reference picture, default is 0
[h264 @ 0x55a925e3ba80] decode_slice_header error
[h264 @ 0x55a925e3ba80] QP 4294967066 out of range
[h264 @ 0x55a925e3ba80] decode_slice_header error
[h264 @ 0x55a925e3ba80] no frame!
[h264 @ 0x55a925e3ba80] QP 341 out of range
[h264 @ 0x55a925e3ba80] decode_slice_header error
I am confused about the difference between MP4 and RTMP of how to use FFmpeg C-API to produce a correct output stream format.
Besides, I also wanna learn how to convert video and audio streams into other formats using FFmpeg C-api, such as flv, ts, rtsp, etc.
Code to reproduce the problem:
I have also put the project files (code, video files) on Github.
The C code shown below can be specifically found at Main.c which is a minimum version to reproduce. It can be reproduced and run successfully.
So how to make this code output to RTMP without getting issue of an unplayable video?
#include <stdio.h>
#include "libavformat/avformat.h"
int main()
{
int ret = 0; int err;
//Open input file
char filename[] = "juren-30s.mp4";
AVFormatContext *fmt_ctx = avformat_alloc_context();
if (!fmt_ctx) {
printf("error code %d \n",AVERROR(ENOMEM));
return ENOMEM;
}
if((err = avformat_open_input(&fmt_ctx, filename,NULL,NULL)) < 0){
printf("can not open file %d \n",err);
return err;
}
//Open the decoder
AVCodecContext *avctx = avcodec_alloc_context3(NULL);
ret = avcodec_parameters_to_context(avctx, fmt_ctx->streams[0]->codecpar);
if (ret < 0){
printf("error code %d \n",ret);
return ret;
}
AVCodec *codec = avcodec_find_decoder(avctx->codec_id);
if ((ret = avcodec_open2(avctx, codec, NULL)) < 0) {
printf("open codec faile %d \n",ret);
return ret;
}
//Open the output file container
char filename_out[] = "juren-30s-5.mp4";
AVFormatContext *fmt_ctx_out = NULL;
err = avformat_alloc_output_context2(&fmt_ctx_out, NULL, NULL, filename_out);
if (!fmt_ctx_out) {
printf("error code %d \n",AVERROR(ENOMEM));
return ENOMEM;
}
//Add all the way to the container context
AVStream *st = avformat_new_stream(fmt_ctx_out, NULL);
st->time_base = fmt_ctx->streams[0]->time_base;
AVCodecContext *enc_ctx = NULL;
AVPacket *pt = av_packet_alloc();
AVFrame *frame = av_frame_alloc();
AVPacket *pkt_out = av_packet_alloc();
int frame_num = 0; int read_end = 0;
for(;;){
if( 1 == read_end ){ break;}
ret = av_read_frame(fmt_ctx, pkt);
//Skip and do not process audio packets
if( 1 == pkt->stream_index ){
av_packet_unref(pt);
continue;
}
if ( AVERROR_EOF == ret) {
//After reading the file, the data and size of pkt should be null at this time
avcodec_send_packet(avctx, NULL);
}else {
if( 0 != ret){
printf("read error code %d \n",ret);
return ENOMEM;
}else{
retry:
if (avcodec_send_packet(avctx, pkt) == AVERROR(EAGAIN)) {
printf("Receive_frame and send_packet both returned EAGAIN, which is an API violation.\n");
//Here you can consider sleeping for 0.1 seconds and returning EAGAIN. This is usually because there is a bug in ffmpeg's internal API.
goto retry;
}
//Release the encoded data in pkt
av_packet_unref(pt);
}
}
//The loop keeps reading data from the decoder until there is no more data to read.
for(;;){
//Read AVFrame
ret = avcodec_receive_frame(avctx, frame);
/* Release the YUV data in the frame,
* Since av_frame_unref is called in the avcodec_receive_frame function, the following code can be commented.
* So we don't need to manually unref this AVFrame
* */
//off_frame_unref(frame);
if( AVERROR(EAGAIN) == ret ){
//Prompt EAGAIN means the decoder needs more AVPackets
//Jump out of the first layer of for and let the decoder get more AVPackets
break;
}else if( AVERROR_EOF == ret ){
/* The prompt AVERROR_EOF means that an AVPacket with both data and size NULL has been sent to the decoder before.
* Sending NULL AVPacket prompts the decoder to flush out all cached frames.
* Usually a NULL AVPacket is sent only after reading the input file, or when another video stream needs to be decoded with an existing decoder.
*
* */
/* Send null AVFrame to the encoder and let the encoder flush out the remaining data.
* */
ret = avcodec_send_frame(enc_ctx, NULL);
for(;;){
ret = avcodec_receive_packet(enc_ctx, pkt_out);
//It is impossible to return EAGAIN here, if there is any, exit directly.
if (ret == AVERROR(EAGAIN)){
printf("avcodec_receive_packet error code %d \n",ret);
return ret;
}
if ( AVERROR_EOF == ret ){ break; }
//Encode the AVPacket, print some information first, and then write it to the file.
printf("pkt_out size : %d \n",pkt_out->size);
//Set the stream_index of AVPacket so that you know which stream it is.
pkt_out->stream_index = st->index;
//Convert the time base of AVPacket to the time base of the output stream.
pkt_out->pts = av_rescale_q_rnd(pkt_out->pts, fmt_ctx->streams[0]->time_base, st->time_base, AV_ROUND_NEAR_INF|AV_ROUND_PASS_MINMAX);
pkt_out->dts = av_rescale_q_rnd(pkt_out->dts, fmt_ctx->streams[0]->time_base, st->time_base, AV_ROUND_NEAR_INF|AV_ROUND_PASS_MINMAX);
pkt_out->duration = av_rescale_q_rnd(pkt_out->duration, fmt_ctx->streams[0]->time_base, st->time_base, AV_ROUND_NEAR_INF|AV_ROUND_PASS_MINMAX);
ret = av_interleaved_write_frame(fmt_ctx_out, pkt_out);
if (ret < 0) {
printf("av_interleaved_write_frame faile %d \n",ret);
return ret;
}
av_packet_unref(pt_out);
}
av_write_trailer(fmt_ctx_out);
//Jump out of the second layer of for, the file has been decoded.
read_end = 1;
break;
}else if( ret >= 0 ){
//Only when a frame is decoded can the encoder be initialized.
if( NULL == enc_ctx ){
//Open the encoder and set encoding information.
AVCodec *encode = avcodec_find_encoder(AV_CODEC_ID_H264);
enc_ctx = avcodec_alloc_context3(encode);
enc_ctx->codec_type = AVMEDIA_TYPE_VIDEO;
enc_ctx->bit_rate = 400000;
enc_ctx->framerate = avctx->framerate;
enc_ctx->gop_size = 30;
enc_ctx->max_b_frames = 10;
enc_ctx->profile = FF_PROFILE_H264_MAIN;
/*
* In fact, the following information is also available in the container. You can also open the encoder directly in the container at the beginning.
* I took these encoder parameters from AVFrame because the difference in the container is final.
* Because the AVFrame you decoded may go through a filter, the information will be transformed after the filter, but this article does not use filters.
*/
//The time base of the encoder should be the time base of AVFrame, because AVFrame is the input. The time base of AVFrame is the time base of the stream.
enc_ctx->time_base = fmt_ctx->streams[0]->time_base;
enc_ctx->width = fmt_ctx->streams[0]->codecpar->width;
enc_ctx->height = fmt_ctx->streams[0]->codecpar->height;
enc_ctx->sample_aspect_ratio = st->sample_aspect_ratio = frame->sample_aspect_ratio;
enc_ctx->pix_fmt = frame->format;
enc_ctx->color_range = frame->color_range;
enc_ctx->color_primaries = frame->color_primaries;
enc_ctx->color_trc = frame->color_trc;
enc_ctx->colorspace = frame->colorspace;
enc_ctx->chroma_sample_location = frame->chroma_location;
/* Note that the value of this field_order is different for different videos. I have written it here.
* Because the video in this article is AV_FIELD_PROGRESSIVE
* The production environment needs to process different videos
*/
enc_ctx->field_order = AV_FIELD_PROGRESSIVE;
/* Now we need to copy the encoder parameters to the stream. When decoding, assign parameters from the stream to the decoder.
* Now let’s do it in reverse.
* */
ret = avcodec_parameters_from_context(st->codecpar,enc_ctx);
if (ret < 0){
printf("error code %d \n",ret);
return ret;
}
if ((ret = avcodec_open2(enc_ctx, encode, NULL)) < 0) {
printf("open codec faile %d \n",ret);
return ret;
}
//Formally open the output file
if ((ret = avio_open2(&fmt_ctx_out->pb, filename_out, AVIO_FLAG_WRITE,&fmt_ctx_out->interrupt_callback,NULL)) < 0) {
printf("avio_open2 fail %d \n",ret);
return ret;
}
//Write the file header first.
ret = avformat_write_header(fmt_ctx_out,NULL);
if (ret < 0) {
printf("avformat_write_header fail %d \n",ret);
return ret;
}
}
//Send AVFrame to the encoder, and then continuously read AVPacket
ret = avcodec_send_frame(enc_ctx, frame);
if (ret < 0) {
printf("avcodec_send_frame fail %d \n",ret);
return ret;
}
for(;;){
ret = avcodec_receive_packet(enc_ctx, pkt_out);
if (ret == AVERROR(EAGAIN)){ break; }
if (ret < 0){
printf("avcodec_receive_packet fail %d \n",ret);
return ret;
}
//Encode the AVPacket, print some information first, and then write it to the file.
printf("pkt_out size : %d \n",pkt_out->size);
//Set the stream_index of AVPacket so that you know which stream it is.
pkt_out->stream_index = st->index;
//Convert the time base of AVPacket to the time base of the output stream.
pkt_out->pts = av_rescale_q_rnd(pkt_out->pts, fmt_ctx->streams[0]->time_base, st->time_base, AV_ROUND_NEAR_INF|AV_ROUND_PASS_MINMAX);
pkt_out->dts = av_rescale_q_rnd(pkt_out->dts, fmt_ctx->streams[0]->time_base, st->time_base, AV_ROUND_NEAR_INF|AV_ROUND_PASS_MINMAX);
pkt_out->duration = av_rescale_q_rnd(pkt_out->duration, fmt_ctx->streams[0]->time_base, st->time_base, AV_ROUND_NEAR_INF|AV_ROUND_PASS_MINMAX);
ret = av_interleaved_write_frame(fmt_ctx_out, pkt_out);
if (ret < 0) {
printf("av_interleaved_write_frame faile %d \n",ret);
return ret;
}
av_packet_unref(pt_out);
}
}
else{ printf("other fail \n"); return ret;}
}
}
av_frame_free(&frame); av_packet_free(&pt); av_packet_free(&pkt_out);
//Close the encoder and decoder.
avcodec_close(avctx); avcodec_close(enc_ctx);
//Release container memory.
avformat_free_context(fmt_ctx);
//Must adjust avio_closep, otherwise the data may not be written in, it will be 0kb
avio_closep(&fmt_ctx_out->pb);
avformat_free_context(fmt_ctx_out);
printf("done \n");
return 0;
}
This problem has haunted over my head for about three weeks. I still have no idea where the key bug exists. Really appreciate it if any FFmpeg expert could help me.
I have fixed this problem. It involves with the sequence of steps when pushing an online stream.
correct sequence:
find encoder -> encoder context (set encoder params)-> allocate packet/frame -> format context &streams.