Guo, Yejun [Mon, 21 Oct 2019 12:38:03 +0000 (20:38 +0800)]
dnn: add tf.nn.conv2d support for native model
Unlike other tf.*.conv2d layers, tf.nn.conv2d does not create many
nodes (within a scope) in the graph, it just acts like other layers.
tf.nn.conv2d only creates one node in the graph, and no internal
nodes such as 'kernel' are created.
The format of native model file is also changed, a flag named
has_bias is added, so change the version number.
Signed-off-by: Guo, Yejun <yejun.guo@intel.com> Signed-off-by: Pedro Arthur <bygrandao@gmail.com>
Philip Langdale [Sun, 27 Oct 2019 18:39:08 +0000 (11:39 -0700)]
avformat/hls: Set AVFMT_TS_DISCONT flag on HLS input format
There have been many reports over the years about problems when
taking an HLS stream as input to `ffmpeg` where there are timestamp
discontinuities present. This is explicitly supported in the
HLS spec (EXT-X-DISCONTINUITY) and often used for ad injection.
Various fixes and work-arounds have been proposed over the years,
but one step that seems obvious, even if it's not a complete fix,
is to mark the HLS input format as supporting discontinuities. This
will prevent timestamp fixup logic in ffmpeg.c kicking in that ends
up mangling the timestamps unnecessarily.
I've tested this out with an example provided by Joe Koberg early
last year, and it is sufficient to allow `ffmpeg` to download and
mux the stream correctly. Joe had briefly suggested that other
situations can still be handled incorrectly, but this seems like
a strict improvement.
Peter Ross [Thu, 20 Jun 2019 21:52:31 +0000 (07:52 +1000)]
vp4: prevent unaligned memory access in loop filter
VP4 applies a loop filter during motion compensation, causing the block offset
will often by unaligned. This produces a bus error on some platforms, namely
ARMv7 NEON.
This patch adds a unaligned version of the loop filter function pointer
to VP3DSPContext.
Reported-by: Mike Melanson <mike@multimedia.cx> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
avcodec/wmalosslessdec: Fix some integer anomalies
Fixes: left shift of negative value -341180 Fixes: 18401/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_WMALOSSLESS_fuzzer-5686380134400000 Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: left shift of negative value -1 Fixes: 18397/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_ADPCM_DTK_fuzzer-5675653487132672 Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Limin Wang [Sat, 26 Oct 2019 13:23:24 +0000 (21:23 +0800)]
avfilter/asrc_anoisesrc: change color variable to int
Or it'll cause invalid color and s->filter is NULL.
Please reproduce it with below command on big endian system:
$ ./ffmpeg -f lavfi -i "anoisesrc=d=60:c=1:r=48000" -f s16le -c:a pcm_s16le -f
null -
Segmentation fault (core dumped)
Signed-off-by: Limin Wang <lance.lmwang@gmail.com> Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
avcodec/libvorbisdec: Fix insufficient input checks leading to out of array reads
Fixes: 16144/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_LIBVORBIS_fuzzer-5638618940440576 Fixes: out of array read Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
avcodec/g723_1dec: fix invalid shift with negative sid_gain
Fixes: left shift of negative value -1 Fixes: 18395/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_G723_1_fuzzer-5710313034350592 Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
ManojGuptaBonda [Fri, 25 Oct 2019 05:30:13 +0000 (11:00 +0530)]
Add support for VP9 VDPAU hwaccel decode
Support for VDPAU accelerated VP9 decoding was added with libvdpau-1.3.
Support for the same in ffmpeg is added with this patch. Profiles
related to VDPAU VP9 can be found in latest vdpau.h present in
libvdpau-1.3. DRC clips are not supported yet due to
http://trac.ffmpeg.org/ticket/8068
Add VP9 VDPAU to list of hwaccels and supported formats
Added file vdpau_vp9.c and Modified configure to add VDPAU VP9 support.
Mapped VP9 profiles to VDPAU VP9 profiles. Populated the codec specific
params that need to be passed to VDPAU.
Signed-off-by: Philip Langdale <philipl@overt.org>
04d2540c added intreadwrite.h to avienc.c, although there was (and is)
no need to do so. The inclusion seems to be a mistake as this commit
added a AV_WL32 to avidec.c.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
libavutil/parseutils.h has been included in 22bbd6e8 for av_parse_time()
and the header has not been removed when said function was replaced by
ff_parse_creation_time_metadata() in ea1bf08a.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Jun Li [Sun, 13 Oct 2019 21:14:25 +0000 (14:14 -0700)]
avformat/tee.c: steal bsf option before passing to fifo muxer
Fix #7620
In the case tee muxer with both "bsf" and "use_fifo" parameters
will trigger this bug. Tee muxer will first steal parameters (like "f",
"select"...) and then "use_fifo" will try reading out remaining options
and pass them to fifo as option "format_options".
Current code miss the part of stealing "bsf" options.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> Reviewed-by: Paul B Mahol <onemda@gmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: Timeout (15sec -> 91ms) Fixes: 18353/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_VP5_fuzzer-5704150326706176 Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg Reviewed-by: Peter Ross <pross@xvid.org> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
avcodec/truespeech: Fix integer overflow in truespeech_synth()
Fixes: signed integer overflow: -1801695444 + -830224908 cannot be represented in type 'int' Fixes: 17995/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_TRUESPEECH_fuzzer-5648084880588800 Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
avcodec/interplayacm: Fix integer overflows in juggle()
Fixes: signed integer overflow: -760459023 + -1520918047 cannot be represented in type 'int' Fixes: 17994/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_INTERPLAY_ACM_fuzzer-5647123042795520 Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
avcodec/hcom: Check the root entry and the right entries
Fixes: Segfault Fixes: 17991/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_HCOM_fuzzer-5647235349479424
Also fixes related memleak
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: left shift of negative value -1 Fixes: 17890/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_RALF_fuzzer-5643307467669504 Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
avcodec/takdec: Fix overflow with large sample rates
Fixes: signed integer overflow: 2147483647 + 511 cannot be represented in type 'int' Fixes: 17899/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_TAK_fuzzer-5719753322135552 Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
In order to fix a potential memleak upon failure, 0b8956b2 made sure that
a buffer given by a pointer was freed upon error. But this pointer was
only initialized upon use and in several cases (Clang gives no fewer
than 13 -Wsometimes-uninitialized warnings) this meant that an
uninitialized pointer was used to free a buffer. So initialize the
pointer.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> Signed-off-by: James Almer <jamrial@gmail.com>
Ramiro Polla [Wed, 23 Oct 2019 19:12:32 +0000 (21:12 +0200)]
mpegvideo_enc: add intra_penalty option for p frames
This option allows more control over the use of intra macroblocks in
predictive frames.
By using '-intra_penalty max', intra macroblocks are never used in
predictive frames.
It is useful for glitch artists to generate input material. This option
allows them to split and merge two video files while maintaining fluid
motion from the second video without having intra macroblocks restoring
chunks of the first video.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Extradata is supposed to be padded with AV_INPUT_BUFFER_PADDING_SIZE bytes,
yet the VobSub demuxer used av_strdup for the allocation of extradata.
This has been changed.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Contains renaming of variables (e.g. mkv_write_cues() contained
variables called tracknum that actually contain the index of a track in
s->streams and not the track number (which can differ in case an
explicit dash track number is set)).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> Signed-off-by: James Almer <jamrial@gmail.com>
avformat/matroskaenc: Use more appropriate function name
mkv_start_new_cluster() actually didn't start a new cluster, but ended
the old one instead and emitted a debug message that it had started a
new cluster. This has been changed: The debug message has been moved to
the place that really starts a new cluster and the function has been
renamed to mkv_end_cluster().
Furthermore, without this debug message the function can be used for
flushing.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> Signed-off-by: James Almer <jamrial@gmail.com>
The Matroska muxer groups index entries with the same pts together in
order to save a few bytes. Because of Matroska's variable-length length
fields, mkv_write_cues() does this by first finding out how many index
entries will be grouped together before actually writing them.
Currently, it is asserted at both of these stages that the stream index
of the list of designated index entries is valid. But the second assert
is redundant, because the very same index entries have already been
checked.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> Signed-off-by: James Almer <jamrial@gmail.com>
The Matroska muxer up until now leaked memory in two scenarios:
1. If an error happened during writing the trailer, as
mkv_write_trailer() returned early without cleaning up.
2. If mkv_write_header() indicated success despite an error in the
underlying AVIOContext. In this case avformat_write_header() returned
the IO error and according to the API the caller is not allowed to call
av_write_trailer(), so that no cleanup happened for the allocations made
in mkv_write_header().
This has been fixed by using a dedicated deinit function.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> Signed-off-by: James Almer <jamrial@gmail.com>