When one merges two AVFilterChannelLayouts structs, there is no need to
allocate a new one. Instead one can reuse one of the two given ones.
If one does this, one also doesn't need to update the references of the
AVFilterChannelLayouts that is reused. Therefore this commit reuses the
structure with the higher refcount.
Reviewed-by: Nicolas George <george@nsup.org>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
The channel layouts accepted by ff_merge_channel_layouts() are of two
types: Ordinary channel layouts and generic channel layouts. These are
layouts that match all layouts with a certain number of channels.
Therefore parsing these channel layouts is not done in one go; instead
first the intersection of the ordinary layouts of the first input
list of channel layouts with the ordinary layouts of the second list is
determined, then the intersection of the ordinary layouts of the first
one and the generic layouts of the second one etc. In order to mark the
ordinary channel layouts that have already been matched as used they are
zeroed. The inner loop that does this is as follows:
for (j = 0; j < b->nb_channel_layouts; j++) {
if (a->channel_layouts[i] == b->channel_layouts[j]) {
ret->channel_layouts[ret_nb++] = a->channel_layouts[i];
a->channel_layouts[i] = b->channel_layouts[j] = 0;
}
}
(Here ret->channel_layouts is the array containing the intersection of
the two input arrays.)
Yet the problem with this code is that after a match has been found, the
loop continues the search with the new value a->channel_layouts[i].
The intention of zeroing these elements was to make sure that elements
already paired at this stage are ignored later. And while they are indeed
ignored when pairing ordinary and generic channel layouts later, it has
the exact opposite effect when pairing ordinary channel layouts.
To see this consider the channel layouts A B C D E and E D C B A. In the
first round, A and A will be paired and added to ret->channel_layouts.
In the second round, the input arrays are 0 B C D E and E D C B 0.
At first B and B will be matched and zeroed, but after doing so matching
continues, but this time it will search for 0, which will match with the
last entry of the second array. ret->channel_layouts now contains A B 0.
In the third round, C 0 0 will be added to ret->channel_layouts etc.
This gives a quadratic amount of elements, yet the amount of elements
allocated for said array is only the sum of the sizes of a and b.
This issue can e.g. be reproduced by
ffmpeg -f lavfi -i anullsrc=cl=7.1 \
-af 'aformat=cl=mono|stereo|2.1|3.0|4.0,aformat=cl=4.0|3.0|2.1|stereo|mono' \
-f null -
The fix is easy: break out of the inner loop after having found a match.
Reviewed-by: Nicolas George <george@nsup.org>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
This reverts commit f156f4ab23.
The checks added by said commit are nonsense because they did not help
in case ff_merge_samplerates() or ff_merge_formats() returned NULL
while freeing one of its arguments: Said freeing does not change
the local variables of can_merge_formats().
Reviewed-by: Nicolas George <george@nsup.org>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Now that the output's refs-array is only allocated once, it is NULL in
any error case and therefore needn't be freed at all; Instead an
av_assert1() has been added to guarantee it to be NULL.
Furthermore, it is unnecessary to av_freep(&ptr) when ptr == NULL.
Reviewed-by: Nicolas George <george@nsup.org>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
ff_merge_formats(), ff_merge_samplerates() and ff_merge_channel_layouts()
share common semantics: If merging succeeds, a non-NULL pointer is
returned and both input lists (of type AVFilterFormats resp.
AVFilterChannelLayouts) are to be treated as if they had been freed;
the owners of the input parameters (if any) become owners of the
returned list. If merging does not succeed, NULL is returned and both
input lists are supposed to be unchanged.
The problem is that the functions did not abide by these semantics:
In case of reallocation failure, it is possible for these functions
to return NULL after having already freed one of the two input list.
This happens because sometimes the refs-array of the destined output
gets reallocated twice to its final size and if the second of these
reallocations fails, the first of the two inputs has already been freed
and its refs updated to point to the destined output which in this case
will be freed immediately so that all of the already updated pointers
are now dangling. This leads to use-after-frees and memory corruptions
lateron (when these owners get cleaned up, the lists they own get
unreferenced). Should the input lists don't have owners at all, the
caller (namely can_merge_formats() in avfiltergraph.c) thinks that both
the input lists are unchanged and need to be freed, leading to a double
free.
The solution to this is simple: Don't reallocate twice; do it just once.
This also saves a reallocation.
This commit fixes the issue behind Coverity issue #1452636. It might
also make Coverity realize that the issue has been fixed.
Reviewed-by: Nicolas George <george@nsup.org>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Despite its name, this function is not part of the public API, as
formats.h, the header containing its declaration, is a private header.
The formats API was once public API, but that changed long ago
(b74a1da49d, the commit scheduling it to
become private, is from 2012). That avfilter_make_format64_list() was
forgotten is probably a result of the confusion resulting from the
libav-ffmpeg split.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
It is unused since 8cbb055760 and it
actually coincides with avfilter_make_format64_list().
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
different backend might need different options for a better performance,
so, add the parameter into dnn interface, as a preparation.
Signed-off-by: Guo, Yejun <yejun.guo@intel.com>
flush_put_bits() already fills the bitstream with zeroes, so it is
unnecessary to align the bitstream before.
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
fix the command ffmpeg -h filter=setpts/asetpts both dump the expr
option with "FVA" flags.
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>
Or it'll cause null pointer dereference if size < sizeof(uint32_t), also
in case tc[0] > 3, the code will report error directly.
Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
Important part of this algorithm is the double threshold step: pixels
above "high" threshold being kept, pixels below "low" threshold dropped,
pixels in between (weak edges) are kept if they are neighboring "high"
pixels.
The weak edge check uses a neighboring context and should not be applied
on the plane's border. The condition was incorrect and has been fixed in
the commit.
Signed-off-by: Andriy Gelman <andriy.gelman@gmail.com>
Reviewed-by: Andriy Gelman <andriy.gelman@gmail.com>
We should not silently allocate an incorrect sized buffer.
Fixes trac issue #8718.
Signed-off-by: Reimar Döffinger <Reimar.Doeffinger@gmx.de>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Reviewed-by: Guo, Yejun <yejun.guo@intel.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>
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>
Note for info level, one extra \n will be print after the log.
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
Currently, the zoompan filter exposes a 'time' variable (missing from docs) for use in
the 'zoom', 'x', and 'y' expressions. This variable is perhaps better named
'out_time' as it represents the timestamp in seconds of each output frame
produced by zoompan. This patch adds aliases 'out_time' and 'ot' for 'time'.
This patch also adds an 'in_time' (alias 'it') variable that provides access
to the timestamp in seconds of each input frame to the zoompan filter.
This helps to design zoompan filters that depend on the input video timestamps.
For example, it makes it easy to zoom in instantly for only some portion of a video.
Both the 'out_time' and 'in_time' variables have been added in the documentation
for zoompan.
Example usage of 'in_time' in the zoompan filter to zoom in 2x for the
first second of the input video and 1x for the rest:
zoompan=z='if(between(in_time,0,1),2,1):d=1'
V2: Fix zoompan filter documentation stating that the time variable
would be NAN if the input timestamp is unknown.
V3: Add 'it' alias for 'in_time. Add 'out_time' and 'ot' aliases for 'time'.
Minor corrections to zoompan docs.
Signed-off-by: exwm <thighsman@protonmail.com>
Guo, Yejun
Andreas Rheinhardt
Mingyu Yin
Paul B Mahol
Ting Fu
leozhang
James Almer
Jun Zhao
Ben Clayton
Limin Wang
Valery Kot
Reimar Döffinger
exwm