Decoding H.264 to YUV drops one frame

[Takis Issaris]

For the attached samples encoded with x264, FFmpeg drops one frame when decoding them to YUV420.

I used the reference decoder (JM 17.2) to verify that the bitstream indeed did contain one more frame than FFmpeg decoded.

./ldecod.dbg.exe -i output3.h264
Setting Default Parameters...
Parsing Configfile decoder.cfg
...............

---

---

Input H.264 bitstream : /tmp/output3.h264
Output decoded YUV : test_dec.yuv
Input reference file : test_rec.yuv

---

POC must = frame# or field# for SNRs to be correct

---

Frame POC Pic# QP SnrY SnrU SnrV Y:U:V Time(ms)

---

Input reference file : test_rec.yuv does not exist

SNR values are not available

(fn=0 T: poc=0 B: poc=0 F: poc=0 G: poc=0) ref (3) view_id (0)
00000(IDR) 0 0 28 4:2:0 86
(fn=0 T: poc=0 B: poc=0 F: poc=0 G: poc=0) ref (3) view_id (0)
(fn=1 T: poc=4 B: poc=4 F: poc=4 G: poc=4) ref (3) view_id (0)
00002( P ) 4 1 28 4:2:0 13
(fn=0 T: poc=0 B: poc=0 F: poc=0 G: poc=0) ref (3) view_id (0)
(fn=1 T: poc=4 B: poc=4 F: poc=4 G: poc=4) ref (3) view_id (0)
(fn=2 T: poc=2 B: poc=2 F: poc=2 G: poc=2)
00001( b ) 2 2 28 4:2:0 10

---

SNR Y(dB) : 0.00
SNR U(dB) : 0.00
SNR V(dB) : 0.00
Total decoding time : 0.111 sec (27.027 fps)[3 frm/111 ms]

---

Exit JM 17 (FRExt) decoder, ver 17.2
Output status file : log.dec

pissaris@takis-Latitude-E6520:/usr/local/src/jm-17.2/bin$ ffmpeg -y -i /tmp/output3.h264 test.yuv
ffmpeg version N-37401-g151ecc2 Copyright (c) 2000-2012 the FFmpeg developers

built on Jan 31 2012 15:39:01 with clang 3.0 (tags/RELEASE_30/final)
configuration: --cc=/opt/clang/bin/clang
libavutil 51. 36.100 / 51. 36.100
libavcodec 54. 0.102 / 54. 0.102
libavformat 54. 0.100 / 54. 0.100
libavdevice 53. 4.100 / 53. 4.100
libavfilter 2. 60.100 / 2. 60.100
libswscale 2. 1.100 / 2. 1.100
libswresample 0. 6.100 / 0. 6.100

[h264 @ 0x29c83a0] decoding for stream 0 failed
[h264 @ 0x29c83a0] Estimating duration from bitrate, this may be inaccurate
Input #0, h264, from '/tmp/output3.h264':

Duration: N/A, bitrate: N/A

Stream #0:0: Video: h264 (High), yuv420p, 352x288, 25 fps, 25 tbr, 1200k tbn, 50 tbc

[buffer @ 0x29d08e0] w:352 h:288 pixfmt:yuv420p tb:1/1000000 sar:0/1 sws_param:
Output #0, rawvideo, to 'test.yuv':

Metadata:

encoder : Lavf54.0.100
Stream #0:0: Video: rawvideo (I420 / 0x30323449), yuv420p, 352x288, q=2-31, 200 kb/s, 90k tbn, 25 tbc

Stream mapping:

Stream #0:0 -> #0:0 (h264 -> rawvideo)

Press [q] to stop, ? for help
frame= 2 fps= 0 q=0.0 Lsize= 297kB time=00:00:00.08 bitrate=30412.8kbits/s dup=0 drop=1
video:297kB audio:0kB global headers:0kB muxing overhead 0.000000%

[Takis Issaris]

H.264 bitstream containing 3 frames

[Takis Issaris]

H.264 bitstream containing 4 frames

[Takis Issaris]

The output4.h264 file contains 4 frames, and similarly FFmpeg decodes only three of them.

[Takis Issaris]

Oh and the same happens with GCC, so it is not clang related.

./ffmpeg -y -i /tmp/output3.h264 test.yuv
ffmpeg version N-37401-g151ecc2 Copyright (c) 2000-2012 the FFmpeg developers

built on Jan 31 2012 15:51:38 with gcc 4.6.1
configuration: --enable-avfilter --enable-frei0r --enable-gpl --enable-libfaac --enable-libmp3lame --enable-libschroedinger --enable-libspeex --enable-libtheora --enable-libvorbis --enable-libvpx --enable-libx264 --enable-libxvid --enable-nonfree --enable-pic --enable-pthreads --disable-shared --enable-static --enable-version3 --enable-x11grab --disable-optimizations
libavutil 51. 36.100 / 51. 36.100
libavcodec 54. 0.102 / 54. 0.102
libavformat 54. 0.100 / 54. 0.100
libavdevice 53. 4.100 / 53. 4.100
libavfilter 2. 60.100 / 2. 60.100
libswscale 2. 1.100 / 2. 1.100
libswresample 0. 6.100 / 0. 6.100
libpostproc 52. 0.100 / 52. 0.100

[h264 @ 0x2de83a0] decoding for stream 0 failed
[h264 @ 0x2de83a0] Estimating duration from bitrate, this may be inaccurate
Input #0, h264, from '/tmp/output3.h264':

Duration: N/A, bitrate: N/A

Stream #0:0: Video: h264 (High), yuv420p, 352x288, 25 fps, 25 tbr, 1200k tbn, 50 tbc

[buffer @ 0x2df08e0] w:352 h:288 pixfmt:yuv420p tb:1/1000000 sar:0/1 sws_param:
Output #0, rawvideo, to 'test.yuv':

Metadata:

encoder : Lavf54.0.100
Stream #0:0: Video: rawvideo (I420 / 0x30323449), yuv420p, 352x288, q=2-31, 200 kb/s, 90k tbn, 25 tbc

Stream mapping:

Stream #0:0 -> #0:0 (h264 -> rawvideo)

Press [q] to stop, ? for help
frame= 2 fps= 0 q=0.0 Lsize= 297kB time=00:00:00.08 bitrate=30412.8kbits/s dup=0 drop=1
video:297kB audio:0kB global headers:0kB muxing overhead 0.000000%

[Carl Eugen Hoyos]

Did you test -vsync 0?
It should allow you to see all encoded frames.

[Takis Issaris]

It does work with vsync 0, but I do not understand why it should drop frames by default. I'd say it should be assumed that there is no difference in framerate between input and output as none was mentioned on the commandline and the outputformat does not force a specific framerate.

[Takis Issaris]

But thanks for mentioning a workaround! :)

[Michael Niedermayer]

It seems this problem has been fixed (possibly by reimars vsync / encoder pts related changes, possibly by something else ...)
as i do get 3 and 4 frames

if iam wrong, please reopen