I'm trying to decode my Hevc file in c++ with using FFmpeg. I used Hevc decoder and try to save the frames in ppm format(Almost the whole source code comes from FFmpeg example [decode_video.c] https://ffmpeg.org/doxygen/trunk/decode_video_8c-example.html, what's new is the conversion from yuv to rgb). My Hevc file has 677 frames, which i checked with ffprobe in command window. But i any got 676 frames with my project. Also i have checked with other Hevc files, the results are same, i got always one frame less. I also tried another FFmpeg example [demuxing_decoding.c] (https://ffmpeg.org/doxygen/trunk/demuxing_decoding_8c-example.html), the result is same, one frame less...
That seems to just happy with H265 and H264 files, is it a bug of FFmpeg?
Can anybody help me, i post my code here. Sorry, don't know how to attach my project and test files. Thanks a lot!
Best regards, Ivan
#include <iostream>
extern "C"
{
#include "../Headers/libavcodec/avcodec.h"
#include "../Headers/libavformat/avformat.h"
#include "../Headers/libswscale/swscale.h"
}
#define INBUF_SIZE 4096
//Save RGB image as PPM file format
static void ppm_save(char* filename, AVFrame* frame)
{
FILE* file;
int i;
fopen_s(&file, filename, "wb");
fprintf(file, "P6\n%d %d\n%d\n", frame->width, frame->height, 255);
for (i = 0; i < frame->height; i++)
fwrite(frame->data[0] + i * frame->linesize[0], 1, frame->width * 3, file);
fclose(file);
}
void decode(AVCodecContext* dec_ctx, AVFrame* frame, AVPacket* pkt, const char* outfilePrefix)
{
char buf[1024];
int ret;
ret = avcodec_send_packet(dec_ctx, pkt);
if (ret < 0) {
fprintf(stderr, "Error sending a packet for decoding\n");
exit(1);
}
int sts;
////////////////////////////////////////////////////////////////////////////
//Create SWS Context for converting from decode pixel format (like YUV420) to RGB
struct SwsContext* sws_ctx = NULL;
sws_ctx = sws_getContext(dec_ctx->width,
dec_ctx->height,
dec_ctx->pix_fmt,
dec_ctx->width,
dec_ctx->height,
AV_PIX_FMT_RGB24,
SWS_BICUBIC,
NULL,
NULL,
NULL);
if (sws_ctx == nullptr)
{
return; //Error!
}
//Allocate frame for storing image converted to RGB.
AVFrame* pRGBFrame = av_frame_alloc();
pRGBFrame->format = AV_PIX_FMT_RGB24;
pRGBFrame->width = dec_ctx->width;
pRGBFrame->height = dec_ctx->height;
sts = av_frame_get_buffer(pRGBFrame, 0);
if (sts < 0)
{
goto free;
//return; //Error!
}
while (ret >= 0)
{
ret = avcodec_receive_frame(dec_ctx, frame);
if (ret == AVERROR(EAGAIN) || ret == AVERROR_EOF)
goto free;
//return;
else if (ret < 0) {
fprintf(stderr, "Error during decoding\n");
exit(1);
}
printf("saving frame %3d\n", dec_ctx->frame_number);//
fflush(stdout);
//////////////////////////////////////////////////////////////////////////
//Convert from input format (e.g YUV420) to RGB and save to PPM:
sts = sws_scale(sws_ctx, //struct SwsContext* c,
frame->data, //const uint8_t* const srcSlice[],
frame->linesize, //const int srcStride[],
0, //int srcSliceY,
frame->height, //int srcSliceH,
pRGBFrame->data, //uint8_t* const dst[],
pRGBFrame->linesize); //const int dstStride[]);
snprintf(buf, sizeof(buf), "%s-%d.ppm", outfilePrefix, dec_ctx->frame_number);
ppm_save(buf, pRGBFrame);
}
free:
//Free
////////////////////////////////////////////////////////////////////////////
sws_freeContext(sws_ctx);
av_frame_free(&pRGBFrame);
}
int main()
{
const char* filename, * outfilePrefix, * seqfilename;
const AVCodec* codec;
AVCodecParserContext* parser;
AVCodecContext* codecContext = NULL;
FILE* file;
AVFrame* frame;
uint8_t inbuf[INBUF_SIZE + AV_INPUT_BUFFER_PADDING_SIZE];
uint8_t* data;
size_t data_size;
int ret;
AVPacket* pkt;
#ifdef _DEBUG
filename = "D:\\TestFiles\\sample_1280x720.hevc";
outfilePrefix = "D:\\TestFiles\\sample_1280x720_output\\output";
#else
if (argc <= 2) {
fprintf(stderr, "Usage: %s <input file> <output file>\n"
"And check your input file is encoded by mpeg1video please.\n", argv[0]);
exit(0);
}
filename = argv[1];
outfilePrefix = argv[2];
#endif
pkt = av_packet_alloc();
if (!pkt)
exit(1);
/* set end of buffer to 0 (this ensures that no overreading happens for damaged MPEG streams) */
memset(inbuf + INBUF_SIZE, 0, AV_INPUT_BUFFER_PADDING_SIZE);
/* find the HEVC video decoder */
codec = avcodec_find_decoder(AV_CODEC_ID_HEVC);
if (!codec) {
fprintf(stderr, "Codec not found\n");
exit(1);
}
parser = av_parser_init(codec->id);
if (!parser) {
fprintf(stderr, "parser not found\n");
exit(1);
}
codecContext = avcodec_alloc_context3(codec);
if (!codecContext) {
fprintf(stderr, "Could not allocate video codec context\n");
exit(1);
}
/* For some codecs, such as msmpeg4 and mpeg4, width and height
MUST be initialized there because this information is not
available in the bitstream. */
/* open it */
if (avcodec_open2(codecContext, codec, NULL) < 0) {
fprintf(stderr, "Could not open codec\n");
exit(1);
}
fopen_s(&file, filename, "rb");
if (!file) {
fprintf(stderr, "Could not open %s\n", filename);
exit(1);
}
frame = av_frame_alloc();
if (!frame) {
fprintf(stderr, "Could not allocate video frame\n");
exit(1);
}
while (!feof(file)) {
/* read raw data from the input file */
data_size = fread(inbuf, 1, INBUF_SIZE, file);
if (!data_size)
break;
/* use the parser to split the data into frames */
data = inbuf;
while (data_size > 0)
{
ret = av_parser_parse2(parser, codecContext, &pkt->data, &pkt->size,
data, data_size, AV_NOPTS_VALUE, AV_NOPTS_VALUE, 0);
if (ret < 0) {
fprintf(stderr, "Error while parsing\n");
exit(1);
}
data += ret;
data_size -= ret;
if (pkt->size)
decode(codecContext, frame, pkt, outfilePrefix);
}
}
/* flush the decoder */
decode(codecContext, frame, NULL, outfilePrefix);
fclose(file);
av_parser_close(parser);
avcodec_free_context(&codecContext);
av_frame_free(&frame);
av_packet_free(&pkt);
}
the problem is that you're not calling av_parser_parse2() with data_size=0 to signal EOF. See the API docs:
buf_size: input length, to signal EOF, this should be 0 (so that the last frame can be output).
Without that call, one frame will be cached in the parser, and that's the one missing in your output.
[edit]
To be clear, I acknowledge that you copied the example code in the API docs correctly:
[..]
while(in_len){
len = av_parser_parse2(myparser, AVCodecContext, &data, &size,
in_data, in_len,
pts, dts, pos);
[..]
However, that code is unfortunately incomplete. If you look at the relevant usage of that code in demux.c, you'll see that explicit flush is required:
[..]
1134 while (size > 0 || (flush && got_output)) {
1135 int64_t next_pts = pkt->pts;
1136 int64_t next_dts = pkt->dts;
1137 int len;
1138
1139 len = av_parser_parse2(sti->parser, sti->avctx,
1140 &out_pkt->data, &out_pkt->size, data, size,
1141 pkt->pts, pkt->dts, pkt->pos);
[..]