Instead use ffio_read_size to read data into a buffer. Also check that
the desired size was actually successfully read and combine the check
with the check for reading the extradata.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Allocating two arrays with the same number of elements together
simplifies freeing them.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
A Smacker file can contain up to seven audio tracks. Up until now,
the pts for the i. audio packet contained in a Smacker frame was
simply the end timestamp of the last i. audio packet contained in
an earlier Smacker frame.
The problem with this is that a Smacker stream need not contain data in
every Smacker frame and so the current i. audio packet present may come
from a different underlying stream than the last i. audio packet
contained in an earlier frame.
The sample hypnotix.smk* exhibits this. It has three audio tracks and
the first of the three has a longer first packet, so that the audio for
the first track is contained in only 235 packets contained in the first
235 Smacker frames; the end timestamp of this track is 166696 (about 7.56s
at a timebase of 1/22050); the other two audio tracks both have 253 packets
contained in the first 253 Smacker frames. Up until now, the 236th
packet of the second track being the first audio packet in the 236th
Smacker frame would get the end timestamp of the last first audio packet
from the last Smacker frame containing a first audio packet and said
last audio packet is the first audio packet from the 235th Smacker frame
from the first audio track, so that the timestamp is 166696. In contrast,
the 236th packet from the third track (whose packets contain the same number
of samples as the packets from the second track) has a timestamp of
156116 (because its timestamp is derived from the end timestamp of the
235th packet of the second audio track). In the end, the second track
ended up being 177360/22050 s = 8.044s long; in contrast, the third
track was 166780/22050 s = 7.56s long which also coincided with the
video.
This commit fixes this by not using timestamps from other tracks for
a packet's pts.
*: https://samples.ffmpeg.org/game-formats/smacker/wetlands/hypnotix.smk
Reviewed-by: Timotej Lazar <timotej.lazar@araneo.si>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
The layout of a Smacker frame is as follows: For some frames, the
beginning of the frame contained a palette for the video stream; then
there are potentially several audio frames, followed by the data for the
video stream.
The Smacker demuxer used to read the palette, then cache every audio frame
into a buffer (that gets reallocated to the desired size every time a
frame is read into this buffer), then read and return the video frame
(together with the palette). The cached audio frames are then returned
by copying the data into freshly allocated buffers; if there are none
left, the next frame is read.
This commit changes this: At the beginning of a frame, the palette is
read and cached as now. But audio frames are no longer cached at all;
they are returned immediately. This gets rid of copying and also allows
to remove the code for the buffer-to-AVStream correspondence.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
The first four bytes of smacker audio are supposed to contain the number
of samples, so treat audio frames smaller than that as invalid.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
When reading a new frame, the Smacker demuxer seeks to the next frame
position where it excepts the next frame; then it (potentially) reads
the palette, the audio packets associated with the frame and finally the
actual video frame. It is only at the end that the frame counter as well
as the position where the next frame is expected get updated.
This has a downside: If e.g. invalid data is encountered when reading
the palette, the demuxer returns immediately (with an error) and if the
caller calls av_read_frame again, the demuxer seeks to the position where
it already was, reads the very same palette data again and therefore will
return an error again. If the caller calls av_read_frame repeatedly
(say, until a packet is received or until EOF), this meight become an
infinite loop.
This could also happen if e.g. the size of one of the audio frames was
invalid or if the frame size was gigantic.
This commit changes this by skipping a frame if it turns out to be
invalid or an error happens otherwise. This ensures that EOF will be
returned eventually in the above scenario.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
The Smacker demuxer buffers audio packets before it outputs them, but it
increments the counter of buffered packets prematurely: If allocating
the audio buffer fails, an error (most likely AVERROR(ENOMEM)) is returned.
If the caller decides to call av_read_frame() again, the next call will
take the codepath for returning already buffered audio packets and it
will fail (because the buffer that ought to be allocated isn't) without
decrementing the number of supposedly buffered audio packets (it doesn't
matter whether there would be enough memory available in subsequent calls).
Depending on the caller's behaviour this is potentially an infinite loop.
This commit fixes this by only incrementing the number of buffered audio
packets after having successfully read them and unconditionally reducing
said number when outputting one of them. It also changes the semantics
of the curstream variable: It is now the number of currently buffered
audio packets whereas it used to be the index of the last audio stream
to be read. (Index refers to the index in the array of buffers, not to
the stream index.)
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
This useful, because by ffprobe's very nature, you use it to probe
a file and find out what it is. Requiring every format private option
to be known to the demuxer forces one to run ffprobe twice, if one
wants to use ffprobe in a generic way.
For example, say one wants to probe all user-uploaded files, while
also ignoring edit lists for any MP4s that are uploaded. Currently,
you'd have to run ffprobe twice: once to identify the format, and
once again to actually probe the metadata you want. After this
patch, you could set -ignore_editlist 1 on every call and only
probe once.
Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
Fixes: signed integer overflow: 6500736 * 473 cannot be represented in type 'int'
Fixes: 23259/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_MV30_fuzzer-5179394271477760
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: index 128 out of bounds for type 'float [128]'
Fixes: 23465/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_HCA_fuzzer-5089866596745216
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: out of array access
Fixes: 23589/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_TIFF_fuzzer-5110559589793792.fuzz
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Real files do skip coding 0 bits at the end, thus this kind of check
does not work reliable.
Fixes: Ticket 8770
Fixes: dst-256fs44-6ch-refdstencoder.dff
The samplerate is specified in ISO/IEC 14496-3:2005(E) as one of 3 fixed
values, this also can be used to limit the duration and avoid the timeout
This reverts commit f6df99dba1.
Don't need to do double check by the description of the API.
Reviewed-by: Nicolas George <george@nsup.org>
Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
floating point precision will cause rgb*max generate different value on
x86_32 and x86_64. have pass fate test on x86_32 and x86_64 by using
lrintf to get the nearest integral value for rgb * max before av_clip.
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
This is the only use of 'FontName' with that capitalization, as both
source-code and tests use 'Fontname'. Having consistent capitalization
makes it easier to find the relevant source from the docs.
See these examples for other uses:
libavcodec/ass_split.c:68
tests/ref/fate/sub-cc:9
We can try with the srcnn model from sr filter.
1) get srcnn.pb model file, see filter sr
2) convert srcnn.pb into openvino model with command:
python mo_tf.py --input_model srcnn.pb --data_type=FP32 --input_shape [1,960,1440,1] --keep_shape_ops
See the script at https://github.com/openvinotoolkit/openvino/tree/master/model-optimizer
We'll see srcnn.xml and srcnn.bin at current path, copy them to the
directory where ffmpeg is.
I have also uploaded the model files at https://github.com/guoyejun/dnn_processing/tree/master/models
3) run with openvino backend:
ffmpeg -i input.jpg -vf format=yuv420p,scale=w=iw*2:h=ih*2,dnn_processing=dnn_backend=openvino:model=srcnn.xml:input=x:output=srcnn/Maximum -y srcnn.ov.jpg
(The input.jpg resolution is 720*480)
Also copy the logs on my skylake machine (4 cpus) locally with openvino backend
and tensorflow backend. just for your information.
$ time ./ffmpeg -i 480p.mp4 -vf format=yuv420p,scale=w=iw*2:h=ih*2,dnn_processing=dnn_backend=tensorflow:model=srcnn.pb:input=x:output=y -y srcnn.tf.mp4
…
frame= 343 fps=2.1 q=31.0 Lsize= 2172kB time=00:00:11.76 bitrate=1511.9kbits/s speed=0.0706x
video:1973kB audio:187kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: 0.517637%
[aac @ 0x2f5db80] Qavg: 454.353
real 2m46.781s
user 9m48.590s
sys 0m55.290s
$ time ./ffmpeg -i 480p.mp4 -vf format=yuv420p,scale=w=iw*2:h=ih*2,dnn_processing=dnn_backend=openvino:model=srcnn.xml:input=x:output=srcnn/Maximum -y srcnn.ov.mp4
…
frame= 343 fps=4.0 q=31.0 Lsize= 2172kB time=00:00:11.76 bitrate=1511.9kbits/s speed=0.137x
video:1973kB audio:187kB subtitle:0kB other streams:0kB global headers:0kB muxing overhead: 0.517640%
[aac @ 0x31a9040] Qavg: 454.353
real 1m25.882s
user 5m27.004s
sys 0m0.640s
Signed-off-by: Guo, Yejun <yejun.guo@intel.com>
Signed-off-by: Pedro Arthur <bygrandao@gmail.com>
OpenVINO is a Deep Learning Deployment Toolkit at
https://github.com/openvinotoolkit/openvino, it supports CPU, GPU
and heterogeneous plugins to accelerate deep learning inferencing.
Please refer to https://github.com/openvinotoolkit/openvino/blob/master/build-instruction.md
to build openvino (c library is built at the same time). Please add
option -DENABLE_MKL_DNN=ON for cmake to enable CPU path. The header
files and libraries are installed to /usr/local/deployment_tools/inference_engine/
with default options on my system.
To build FFmpeg with openvion, take my system as an example, run with:
$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/deployment_tools/inference_engine/lib/intel64/:/usr/local/deployment_tools/inference_engine/external/tbb/lib/
$ ../ffmpeg/configure --enable-libopenvino --extra-cflags=-I/usr/local/deployment_tools/inference_engine/include/ --extra-ldflags=-L/usr/local/deployment_tools/inference_engine/lib/intel64
$ make
Here are the features provided by OpenVINO inference engine:
- support more DNN model formats
It supports TensorFlow, Caffe, ONNX, MXNet and Kaldi by converting them
into OpenVINO format with a python script. And torth model
can be first converted into ONNX and then to OpenVINO format.
see the script at https://github.com/openvinotoolkit/openvino/tree/master/model-optimizer/mo.py
which also does some optimization at model level.
- optimize at inference stage
It optimizes for X86 CPUs with SSE, AVX etc.
It also optimizes based on OpenCL for Intel GPUs.
(only Intel GPU supported becuase Intel OpenCL extension is used for optimization)
Signed-off-by: Guo, Yejun <yejun.guo@intel.com>
Signed-off-by: Pedro Arthur <bygrandao@gmail.com>
Saves initialization of an HEVCDecoderConfigurationRecord when
the data is already in ISOBMFF-format or if it is plainly invalid.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
This patch adds the control for enabling rectangular partitions, 1:4/4:1
partitions and AB shape partitions.
Signed-off-by: Wang Cao <wangcao@google.com>
Signed-off-by: James Zern <jzern@google.com>
Fix two cases of memleaks:
1. The leak of dv_demux
2. The leak of dv_fctx upon dv_demux allocate failure
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
The code for demuxing DV audio predates the introduction of refcounted
packets and when the latter was added, changes to the former were
forgotten. This meant that when avpriv_dv_produce_packet initialized the
packet containing the AVBufferRef, the AVBufferRef as well as the
underlying AVBuffer leaked; the actual packet data didn't leak: They
were directly freed, but not via their AVBuffer's free function.
https://samples.ffmpeg.org/ffmpeg-bugs/trac/ticket4671/dir1.tar.bz2
contains samples for this (enable_drefs needs to be enabled for them).
Moreover, errors in avpriv_dv_produce_packet were ignored; this has been
changed, too.
Furthermore, in the hypothetical scenario that the track has a palette,
this would leak, too, so reorder the code so that the palette code
appears after the DV audio code.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Fixes: signed integer overflow: 2147483610 + 52 cannot be represented in type 'int'
Fixes: 23260/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_PBM_fuzzer-5187871274434560
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: signed integer overflow: 8 * 9223372036854774783 cannot be represented in type 'long'
Fixes: 23381/clusterfuzz-testcase-minimized-ffmpeg_DEMUXER_fuzzer-4818340509122560
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: out of array access
Fixes: 23554/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_APNG_fuzzer-4796622520451072.fuzz
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The segments / url can be modified by the io read when reloading
This may be an alternative or additional fix for Ticket8673
as a further alternative the reload stuff could be disabled during
probing
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The encoder has no delayed packets at the end of the encoding
process, so signaling this capability is unnecessary.
This also fixes an assertion failure introduced in 827d6fe73d, as
return values higher than 0 are not expected.
Signed-off-by: James Almer <jamrial@gmail.com>
