Corrupt H264 when streaming H264 from v4l2 device

[noah]

I would like to stream H264 without transcoding on a Raspberry Pi with a Raspberry Pi Camera module. Currently it seems FFmpeg gets confused somewhere after data has been captured from the V4L2 device; the H264 demuxer(?) starts spitting out a bunch of errors which seem to indicate it is being fed junk. Depending on the ffmpeg version, I either get corrupted output or no output at all.

I'm seeing the problem in git master and also verified it to be present in tag n1.2.12 and n2.6.1 in git.

The problem seems to be very similar to what was reported in 2013, and subsequently fixed, here:
​https://trac.ffmpeg.org/ticket/2882 (ffmpeg can't save h264 stream from v4l2 webcam into file without recoding)

Changing AVSTREAM_PARSE_HEADERS to AVSTREAM_PARSE_FULL in libavdevice/v4l2.c:v4l2_read_header() fixes the problem for me, though I'm not sure it's the correct solution.

Steps to reproduce:

1. Get a Raspberry Pi with an attached Raspberry Pi Camera module, load bcm2835_v4l kernel module
2. Build ffmpeg from source
3. Run: ./ffmpeg -report -f v4l2 -input_format h264 -t 2 -i /dev/video0 -vcodec copy -loglevel debug -y foo.mkv

$ ./ffmpeg -report -f v4l2 -input_format h264 -t 2 -i /dev/video0 -vcodec copy -loglevel debug -y foo.mkv
...
Opening an input file: /dev/video0.
[video4linux2,v4l2 @ 0x1aec9e0] fd:4 capabilities:85200005
[video4linux2,v4l2 @ 0x1aec9e0] Current input_channel: 0, input_name: Camera 0, input_std: 0
[video4linux2,v4l2 @ 0x1aec9e0] Querying the device for the current frame size
[video4linux2,v4l2 @ 0x1aec9e0] Setting frame size to 1024x768
[h264 @ 0x1aed230] missing picture in access unit with size 27
[h264 @ 0x1aed230] no frame!
[h264 @ 0x1aed230] missing picture in access unit with size 27
[video4linux2,v4l2 @ 0x1aec9e0] Non-increasing DTS in stream 0: packet 11 with DTS 7463370264, packet 12 with DTS 7462329653
[h264 @ 0x1aed230] missing picture in access unit with size 27
[video4linux2,v4l2 @ 0x1aec9e0] Non-increasing DTS in stream 0: packet 22 with DTS 7463770172, packet 23 with DTS 7462729442
[video4linux2,v4l2 @ 0x1aec9e0] Non-increasing DTS in stream 0: packet 26 with DTS 7462849378, packet 27 with DTS 0
[video4linux2,v4l2 @ 0x1aec9e0] Non-increasing DTS in stream 0: packet 27 with DTS 0, packet 28 with DTS 0
[video4linux2,v4l2 @ 0x1aec9e0] Non-increasing DTS in stream 0: packet 28 with DTS 0, packet 29 with DTS 0
[video4linux2,v4l2 @ 0x1aec9e0] Non-increasing DTS in stream 0: packet 29 with DTS 0, packet 30 with DTS 0
[video4linux2,v4l2 @ 0x1aec9e0] Non-increasing DTS in stream 0: packet 30 with DTS 0, packet 31 with DTS 0
[video4linux2,v4l2 @ 0x1aec9e0] Non-increasing DTS in stream 0: packet 31 with DTS 0, packet 32 with DTS 0
[h264 @ 0x1aed230] missing picture in access unit with size 27
... more messages similar to the three above ...
Input #0, video4linux2,v4l2, from '/dev/video0':
  Duration: N/A, start: -140455148.518520, bitrate: N/A
    Stream #0:0, 125, 1/1000000: Video: h264 (High), 1 reference frame, yuv420p(left), 1024x768, 1/2000000, -4 kb/s, 25 fps, 25 tbr, 10
00k tbn, 2000k tbc
Successfully opened the file.
Parsing a group of options: output file foo.mkv.
Applying option vcodec (force video codec ('copy' to copy stream)) with argument copy.
Successfully parsed a group of options.
Opening an output file: foo.mkv.
Successfully opened the file.
[matroska @ 0x1b68230] Codec for stream 0 does not use global headers but container format requires global headers
Output #0, matroska, to 'foo.mkv':
  Metadata:
    encoder         : Lavf56.37.100
    Stream #0:0, 0, 1/1000: Video: h264, 1 reference frame (H264 / 0x34363248), yuv420p(left), 1024x768 (0x0), 1/1000000, q=2-31, -4 kb
/s, 25 fps, 25 tbr, 1k tbn, 1000k tbc
Stream mapping:
  Stream #0:0 -> #0:0 (copy)
Press [q] to stop, [?] for help

A full log is attached as ffmpeg-v4l2-h264-copy-fails-20150620-010456.log.

Ffprobe gives similar errors:

$ ./ffprobe -report -f v4l2 -input_format h264 /dev/video0
...
[h264 @ 0x2fcee10] missing picture in access unit with size 27
[h264 @ 0x2fcee10] no frame!
[h264 @ 0x2fcee10] missing picture in access unit with size 27
    Last message repeated 10 times
Input #0, video4linux2,v4l2, from '/dev/video0':
  Duration: N/A, start: -140455000.691313, bitrate: N/A
    Stream #0:0: Video: h264 (High), yuv420p, 1024x768, -4 kb/s, 25 fps, 25 tbr, 1000k tbn, 2000k tbc
$ echo $?
130

A full log is attached as ffprobe-v4l2-h264-20150620-005024.log.

What does work, however, is to let ffmpeg stream raw video from the camera (e.g. remove '-input_format h264'). Example:

Opening an input file: /dev/video0.
[video4linux2,v4l2 @ 0x1dbc9e0] fd:4 capabilities:85200005
[video4linux2,v4l2 @ 0x1dbc9e0] Current input_channel: 0, input_name: Camera 0, input_std: 0
[video4linux2,v4l2 @ 0x1dbc9e0] Querying the device for the current frame size
[video4linux2,v4l2 @ 0x1dbc9e0] Setting frame size to 1024x768
[video4linux2,v4l2 @ 0x1dbc9e0] All info found
Input #0, video4linux2,v4l2, from '/dev/video0':
  Duration: N/A, start: 9375.983690, bitrate: 235929 kb/s
    Stream #0:0, 1, 1/1000000: Video: rawvideo, 1 reference frame (I420 / 0x30323449), yuv420p, 1024x768, 1/1000000, 235929 kb/s, 25 fps, 25 tbr, 1000k tbn, 1000k tbc
Successfully opened the file.
Parsing a group of options: output file out.mkv.
Applying option vcodec (force video codec ('copy' to copy stream)) with argument copy.
Successfully parsed a group of options.
Opening an output file: out.mkv.
Successfully opened the file.
[matroska @ 0x1dbdc90] Codec for stream 0 does not use global headers but container format requires global headers
Output #0, matroska, to 'out.mkv':
  Metadata:
    encoder         : Lavf56.37.100
    Stream #0:0, 0, 1/1000: Video: rawvideo, 1 reference frame (I420 / 0x30323449), yuv420p, 1024x768 (0x0), 1/1000000, q=2-31, 235929 kb/s, 25 fps, 25 tbr, 1k tbn, 1000k tbc
Stream mapping:
  Stream #0:0 -> #0:0 (copy)
Press [q] to stop, [?] for help
[matroska @ 0x1dbdc90] Writing block at offset 536, size 1179648, pts 0, dts 0, duration 40, keyframe 1
[matroska @ 0x1dbdc90] Starting new cluster at offset 1180192 bytes, pts 182dts 182
[matroska @ 0x1dbdc90] Writing block at offset 1180207, size 1179648, pts 182, dts 182, duration 40, keyframe 1
[matroska @ 0x1dbdc90] Starting new cluster at offset 2359863 bytes, pts 365dts 365
[matroska @ 0x1dbdc90] Writing block at offset 2359879, size 1179648, pts 365, dts 365, duration 40, keyframe 1
[matroska @ 0x1dbdc90] Starting new cluster at offset 3539535 bytes, pts 547dts 547
[matroska @ 0x1dbdc90] Writing block at offset 3539551, size 1179648, pts 547, dts 547, duration 40, keyframe 1
frame=    4 fps=0.0 q=-1.0 size=    4609kB time=00:00:00.54 bitrate=69019.5kbits/s   
[matroska @ 0x1dbdc90] Starting new cluster at offset 4719207 bytes, pts 729dts 729
[matroska @ 0x1dbdc90] Writing block at offset 4719223, size 1179648, pts 729, dts 729, duration 40, keyframe 1
[matroska @ 0x1dbdc90] Starting new cluster at offset 5898879 bytes, pts 912dts 912
[matroska @ 0x1dbdc90] Writing block at offset 5898895, size 1179648, pts 912, dts 912, duration 40, keyframe 1
...
[matroska @ 0x1dbdc90] Writing block at offset 25953319, size 1179648, pts 4010, dts 4010, duration 40, keyframe 1
[matroska @ 0x1dbdc90] end duration = 4050
frame=   23 fps=5.0 q=-1.0 Lsize=   26497kB time=00:00:04.01 bitrate=54131.5kbits/s
video:26496kB audio:0kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: 0.005621%
Input file #0 (/dev/video0):
  Input stream #0:0 (video): 23 packets read (27131904 bytes);
  Total: 23 packets (27131904 bytes) demuxed
Output file #0 (out.mkv):
  Output stream #0:0 (video): 23 packets muxed (27131904 bytes);
  Total: 23 packets (27131904 bytes) muxed
0 frames successfully decoded, 0 decoding errors
[AVIOContext @ 0x1dc74c0] Statistics: 156 seeks, 957 writeouts

A full log is attached as ffmpeg-v4l2-raw-streaming-works-20150620-013650.log.

Since ticket #2882 was marked as fixed, v4l2_read_header() in libavdevice/v4l2.c contains:

st->need_parsing = AVSTREAM_PARSE_HEADERS;

If I change this to AVSTREAM_PARSE_FULL, as was suggested in an initial patch in attached to the ticket, ffmpeg appears to do better and produces playable output data. However, while it no longer complains about H264, it floods the screen with warnings about 'Non-monotonous DTS in output stream'.

...
Opening an input file: /dev/video0.
[video4linux2,v4l2 @ 0x13cd9e0] fd:4 capabilities:85200005
[video4linux2,v4l2 @ 0x13cd9e0] Current input_channel: 0, input_name: Camera 0, input_std: 0
[video4linux2,v4l2 @ 0x13cd9e0] Querying the device for the current frame size
[video4linux2,v4l2 @ 0x13cd9e0] Setting frame size to 1024x768
[video4linux2,v4l2 @ 0x13cd9e0] v4l2_read_header: turning on AVSTREAM_PARSE_FULL
[video4linux2,v4l2 @ 0x13cd9e0] All info found
Input #0, video4linux2,v4l2, from '/dev/video0':
  Duration: N/A, start: 8047.143224, bitrate: N/A
    Stream #0:0, 8, 1/1000000: Video: h264 (High), 1 reference frame, yuv420p(left), 1024x768, 1/2000000, -4 kb/s, 25 fps, 25 tbr, 1000k tbn, 2000k tbc
Successfully opened the file.
Parsing a group of options: output file foo.mkv.
Applying option vcodec (force video codec ('copy' to copy stream)) with argument copy.
Successfully parsed a group of options.
Opening an output file: foo.mkv.
Successfully opened the file.
[matroska @ 0x144acd0] Codec for stream 0 does not use global headers but container format requires global headers
Output #0, matroska, to 'foo.mkv':
  Metadata:
    encoder         : Lavf56.37.100
    Stream #0:0, 0, 1/1000: Video: h264, 1 reference frame (H264 / 0x34363248), yuv420p(left), 1024x768 (0x0), 1/1000000, q=2-31, -4 kb/s, 25 fps, 25 tbr, 1k tbn, 1000k tbc
Stream mapping:
  Stream #0:0 -> #0:0 (copy)
Press [q] to stop, [?] for help
[matroska @ 0x144acd0] Writing block at offset 562, size 1898, pts 0, dts 0, duration 0, keyframe 1
[matroska @ 0x144acd0] Writing block at offset 2467, size 2579, pts 0, dts 0, duration 0, keyframe 0
[matroska @ 0x144acd0] Writing block at offset 5053, size 2148, pts 40, dts 40, duration 0, keyframe 0
[matroska @ 0x144acd0] Writing block at offset 7208, size 3204, pts 80, dts 80, duration 0, keyframe 0
[matroska @ 0x144acd0] Writing block at offset 10419, size 2867, pts 120, dts 120, duration 0, keyframe 0
[matroska @ 0x144acd0] Writing block at offset 13293, size 2737, pts 160, dts 160, duration 0, keyframe 0
[matroska @ 0x144acd0] Writing block at offset 16037, size 3747, pts 200, dts 200, duration 0, keyframe 0
[matroska @ 0x144acd0] Writing block at offset 19791, size 5135, pts 240, dts 240, duration 0, keyframe 0
[matroska @ 0x144acd0] Writing block at offset 24933, size 7273, pts 280, dts 280, duration 0, keyframe 0
[matroska @ 0x144acd0] Writing block at offset 32213, size 12305, pts 320, dts 320, duration 0, keyframe 0
[matroska @ 0x144acd0] Starting new cluster at offset 44525 bytes, pts 1401dts 1401
[matroska @ 0x144acd0] Writing block at offset 44541, size 9172, pts 1401, dts 1401, duration 0, keyframe 1
[matroska @ 0x144acd0] Non-monotonous DTS in output stream 0:0; previous: 1401, current: 400; changing to 1401. This may result in incorrect timestamps in the output file.
[matroska @ 0x144acd0] Writing block at offset 53720, size 7825, pts 1401, dts 1401, duration 0, keyframe 0
[matroska @ 0x144acd0] Non-monotonous DTS in output stream 0:0; previous: 1401, current: 440; changing to 1401. This may result in incorrect timestamps in the output file.
...
[matroska @ 0x144acd0] end duration = 4998
frame=  102 fps= 28 q=-1.0 Lsize=    1657kB time=00:00:04.99 bitrate=2716.7kbits/s    
video:1656kB audio:0kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: 0.101796%
Input file #0 (/dev/video0):
  Input stream #0:0 (video): 102 packets read (1695552 bytes); 
  Total: 102 packets (1695552 bytes) demuxed
Output file #0 (foo.mkv):
  Output stream #0:0 (video): 102 packets muxed (1695552 bytes); 
  Total: 102 packets (1695552 bytes) muxed
0 frames successfully decoded, 0 decoding errors

A full log is attached as ffmpeg-v4l2-h264-copy-AVSTREAM_PARSE_FULL-ok-20150620-011441.log.

FWIW, I see similar problem if I use .ts instead of .mkv for the output file.

My environment is Raspberry Pi (all models) with attached a Raspberry Pi Camera module. Running Raspbian 2015-05-05 (based on Debian 7.8) with the bcm2835-v4l2 kernel module providing a V4L2 interface to the camera.

$ uname -a
Linux raspberrypi 3.18.11-v7+ #781 SMP PREEMPT Tue Apr 21 18:07:59 BST 2015 armv7l GNU/Linux
$ lsmod |egrep 'v4l2|video'
bcm2835_v4l2           37225  0 
videobuf2_vmalloc       3009  1 bcm2835_v4l2
videobuf2_memops        1552  1 videobuf2_vmalloc
videobuf2_core         33578  1 bcm2835_v4l2
v4l2_common             5077  2 bcm2835_v4l2,videobuf2_core
videodev              122487  3 bcm2835_v4l2,v4l2_common,videobuf2_core
media                  11577  1 videodev

Output from v4l2-ctl with --get-fmt-video, --list-formats-ext and --list-ctrls-menus:

$ v4l2-ctl --get-fmt-video
Format Video Capture:
        Width/Height  : 1024/768
        Pixel Format  : 'H264'
        Field         : None
        Bytes per Line: 1024
        Size Image    : 786432
        Colorspace    : Broadcast NTSC/PAL (SMPTE170M/ITU601)
        Custom Info   : feedcafe

$ v4l2-ctl --list-formats-ext 
ioctl: VIDIOC_ENUM_FMT
	Index       : 0
	Type        : Video Capture
	Pixel Format: 'YU12'
	Name        : 4:2:0, packed YUV
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 1
	Type        : Video Capture
	Pixel Format: 'YUYV'
	Name        : 4:2:2, packed, YUYV
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 2
	Type        : Video Capture
	Pixel Format: 'RGB3'
	Name        : RGB24 (LE)
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 3
	Type        : Video Capture
	Pixel Format: 'JPEG' (compressed)
	Name        : JPEG
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 4
	Type        : Video Capture
	Pixel Format: 'H264' (compressed)
	Name        : H264
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 5
	Type        : Video Capture
	Pixel Format: 'MJPG' (compressed)
	Name        : MJPEG
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 6
	Type        : Video Capture
	Pixel Format: 'YVYU'
	Name        : 4:2:2, packed, YVYU
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 7
	Type        : Video Capture
	Pixel Format: 'VYUY'
	Name        : 4:2:2, packed, VYUY
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 8
	Type        : Video Capture
	Pixel Format: 'UYVY'
	Name        : 4:2:2, packed, UYVY
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 9
	Type        : Video Capture
	Pixel Format: 'NV12'
	Name        : 4:2:0, packed, NV12
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 10
	Type        : Video Capture
	Pixel Format: 'BGR3'
	Name        : RGB24 (BE)
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 11
	Type        : Video Capture
	Pixel Format: 'YV12'
	Name        : 4:2:0, packed YVU
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 12
	Type        : Video Capture
	Pixel Format: 'NV21'
	Name        : 4:2:0, packed, NV21
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

	Index       : 13
	Type        : Video Capture
	Pixel Format: 'BGR4'
	Name        : RGB32 (BE)
		Size: Stepwise 16x16 - 2592x1944 with step 2/2

$ v4l2-ctl --list-ctrls-menus

User Controls

                     brightness (int)    : min=0 max=100 step=1 default=50 value=50 flags=slider
                       contrast (int)    : min=-100 max=100 step=1 default=0 value=0 flags=slider
                     saturation (int)    : min=-100 max=100 step=1 default=0 value=0 flags=slider
                    red_balance (int)    : min=1 max=7999 step=1 default=1000 value=1000 flags=slider
                   blue_balance (int)    : min=1 max=7999 step=1 default=1000 value=1000 flags=slider
                horizontal_flip (bool)   : default=0 value=1
                  vertical_flip (bool)   : default=0 value=1
           power_line_frequency (menu)   : min=0 max=3 default=1 value=1
				0: Disabled
				1: 50 Hz
				2: 60 Hz
				3: Auto
                      sharpness (int)    : min=-100 max=100 step=1 default=0 value=0 flags=slider
                  color_effects (menu)   : min=0 max=15 default=0 value=0
				0: None
				1: Black & White
				2: Sepia
				3: Negative
				4: Emboss
				5: Sketch
				6: Sky Blue
				7: Grass Green
				8: Skin Whiten
				9: Vivid
				10: Aqua
				11: Art Freeze
				12: Silhouette
				13: Solarization
				14: Antique
				15: Set Cb/Cr
                         rotate (int)    : min=0 max=360 step=90 default=0 value=0
             color_effects_cbcr (int)    : min=0 max=65535 step=1 default=32896 value=32896

Codec Controls

             video_bitrate_mode (menu)   : min=0 max=1 default=0 value=0 flags=update
				0: Variable Bitrate
				1: Constant Bitrate
                  video_bitrate (int)    : min=25000 max=25000000 step=25000 default=10000000 value=6000000
         repeat_sequence_header (bool)   : default=0 value=1
            h264_i_frame_period (int)    : min=0 max=2147483647 step=1 default=60 value=10
                     h264_level (menu)   : min=0 max=11 default=11 value=11
				0: 1
				1: 1b
				2: 1.1
				3: 1.2
				4: 1.3
				5: 2
				6: 2.1
				7: 2.2
				8: 3
				9: 3.1
				10: 3.2
				11: 4
                   h264_profile (menu)   : min=0 max=4 default=4 value=4
				0: Baseline
				1: Constrained Baseline
				2: Main
				4: High

Camera Controls

                  auto_exposure (menu)   : min=0 max=3 default=0 value=0
				0: Auto Mode
				1: Manual Mode
         exposure_time_absolute (int)    : min=1 max=10000 step=1 default=1000 value=1000
     exposure_dynamic_framerate (bool)   : default=0 value=0
             auto_exposure_bias (intmenu): min=0 max=24 default=12 value=12
				0: -4000 (0xfffffffffffff060)
				1: -3667 (0xfffffffffffff1ad)
				2: -3333 (0xfffffffffffff2fb)
				3: -3000 (0xfffffffffffff448)
				4: -2667 (0xfffffffffffff595)
				5: -2333 (0xfffffffffffff6e3)
				6: -2000 (0xfffffffffffff830)
				7: -1667 (0xfffffffffffff97d)
				8: -1333 (0xfffffffffffffacb)
				9: -1000 (0xfffffffffffffc18)
				10: -667 (0xfffffffffffffd65)
				11: -333 (0xfffffffffffffeb3)
				12: 0 (0x0)
				13: 333 (0x14d)
				14: 667 (0x29b)
				15: 1000 (0x3e8)
				16: 1333 (0x535)
				17: 1667 (0x683)
				18: 2000 (0x7d0)
				19: 2333 (0x91d)
				20: 2667 (0xa6b)
				21: 3000 (0xbb8)
				22: 3333 (0xd05)
				23: 3667 (0xe53)
				24: 4000 (0xfa0)
      white_balance_auto_preset (menu)   : min=0 max=9 default=1 value=1
				0: Manual
				1: Auto
				2: Incandescent
				3: Fluorescent
				4: Fluorescent H
				5: Horizon
				6: Daylight
				7: Flash
				8: Cloudy
				9: Shade
            image_stabilization (bool)   : default=0 value=0
                iso_sensitivity (intmenu): min=0 max=4 default=0 value=0
				0: 0 (0x0)
				1: 100 (0x64)
				2: 200 (0xc8)
				3: 400 (0x190)
				4: 800 (0x320)
         exposure_metering_mode (menu)   : min=0 max=2 default=0 value=0
				0: Average
				1: Center Weighted
				2: Spot
                     scene_mode (menu)   : min=0 max=13 default=0 value=0
				0: None
				8: Night
				11: Sports

JPEG Compression Controls

            compression_quality (int)    : min=1 max=100 step=1 default=30 value=30

[Carl Eugen Hoyos]

Replying to noah:

Changing AVSTREAM_PARSE_HEADERS to AVSTREAM_PARSE_FULL in libavdevice/v4l2.c:v4l2_read_header() fixes the problem for me, though I'm not sure it's the correct solution.

Could you send a patch made with git format-patch to the development mailing list?

[noah]

Replying to cehoyos:

Replying to noah:

Changing AVSTREAM_PARSE_HEADERS to AVSTREAM_PARSE_FULL in libavdevice/v4l2.c:v4l2_read_header() fixes the problem for me, though I'm not sure it's the correct solution.

Could you send a patch made with git format-patch to the development mailing list?

Thanks for commenting, Carl Eugen. I will consider your suggestion.

I found some interesting comments in libavformat/avformat.h:

enum AVStreamParseType {
    AVSTREAM_PARSE_NONE,
    AVSTREAM_PARSE_FULL,       /**< full parsing and repack */
    AVSTREAM_PARSE_HEADERS,    /**< Only parse headers, do not repack. */
    AVSTREAM_PARSE_TIMESTAMPS, /**< full parsing and interpolation of timestamps for frames not starting on a packet boundary */
    AVSTREAM_PARSE_FULL_ONCE,  /**< full parsing and repack of the first frame only, only implemented for H.264 currently */
    AVSTREAM_PARSE_FULL_RAW=MKTAG(0,'R','A','W'),       /**< full parsing and repack with timestamp and position generation by parser for raw
                                                             this assumes that each packet in the file contains no demuxer level headers and
                                                             just codec level data, otherwise position generation would fail */
};

Of those, all but AVSTREAM_PARSE_NONE and AVSTREAM_PARSE_HEADERS appear to produce video that can be played back in VLC (3.0.0-git). Judging from the notes about repacking, I assume AVSTREAM_PARSE_FULL_ONCE should be the most efficient option of them, no?

I still have problems with non-monotonically increasing DTS and that -t 10 -i /dev/video0 doesn't stop reading after 10 seconds (regardless of which full parsing option I use), but I guess I'll have to ask the Raspberry Pi folks why DTS is wrong in the camera output stream. Example warning messages from the h264 muxer:

[h264 @ 0x148aef0] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 5792923 >= 5716981
[h264 @ 0x148aef0] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 5792923 >= 5756960
[h264 @ 0x148aef0] Application provided invalid, non monotonically increasing dts to muxer 
n stream 0: 10590120 >= 9594933
[h264 @ 0x148aef0] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 10590120 >= 9634912
[h264 @ 0x148aef0] Application provided invalid, non monotonically increasing dts to muxer in stream 0: 10590120 >= 9674891

For the sake of completeness, I'm attaching two new reports where I use AVSTREAM_PARSE_FULL_ONCE:

v4l2-ctl -c video_bitrate=10000000  -c h264_i_frame_period=120  -c repeat_sequence_header=0 -v width=1280,height=720,pixelformat=h264
./ffmpeg-full-once -report -f v4l2 -input_format h264 -framerate 25 -t 10 -i /dev/video0 -c copy -an -loglevel debug  -f h264  -y a-full-once-rsh-0.h264 

and:

v4l2-ctl -c video_bitrate=10000000  -c h264_i_frame_period=120  -c repeat_sequence_header=1 -v width=1280,height=720,pixelformat=h264
./ffmpeg-full-once -report -f v4l2 -input_format h264 -framerate 25 -t 10 -i /dev/video0 -c copy -an -loglevel debug  -f h264  -y a-full-once-rsh-1.h264 

[Carl Eugen Hoyos]

​http://thread.gmane.org/gmane.comp.video.ffmpeg.devel/195493
Could you test the patch (again) and confirm that it fixes an issue for you? I believe you already said so, but I am not a native speaker and others disagree, so an additional confirmation is necessary to fix this ticket.

[noah]

Per the suggestion in the linked thread, I did three more tests (default with AVSTREAM_PARSE_HEADERS, AVSTREAM_PARSE_FULL_ONCE and AVSTREAM_PARSE_FULL) with two different containers (.mov and .nut) and a more recent code base.

I can confirm that the more recent code produce the same results as previously reported.

I've uploaded the resulting files to upload.ffmpeg.org with the following names:

$ ls -1rt |xargs du -sk
9944	ffmpeg-headers.mov
80	ffmpeg-20150722-222055-headers.mov.log
7752	ffmpeg-full-once.mov
48	ffmpeg-20150722-222152-full-once.mov.log
6980	ffmpeg-full.mov
44	ffmpeg-20150722-222221-full.mov.log
4	ffmpeg-headers.nut
16	ffmpeg-20150722-222554-headers.nut.log
6580	ffmpeg-full.nut
40	ffmpeg-20150722-222605-full.nut.log
6716	ffmpeg-full-once.nut
40	ffmpeg-20150722-222620-full-once.nut.log

(I'm sorry I didn't think of including the ticket ID in the filenames before they were uploaded..)

Usage of either AVSTREAM_PARSE_FULL_ONCE or AVSTREAM_PARSE_FULL when capturing in libavdevice/v4l2.c appears to produce at least .mov files that play nicely with Quicktime Player (OSX 10.10) and VLC (v2.2.1).

When attempting to play the the file which was captured with AVSTREAM_PARSE_HEADERS (as is the default in git master) in Quicktime Player, the playback won't start even if I hit the play button. If I manually seek into the middle of the file and start playing from there, all I get is a black screen.

[Carl Eugen Hoyos]

Finally committed as 4c4f14c7179c1c79f71a3036e537b1c1dbc50030
Thank you for the report, the debugging and the testing!