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>
This contains encoder wrappers for H264, HEVC, AAC, AC3 and MP3.
This is based on top of an original patch by wm4
<nfxjfg@googlemail.com>. The original patch supported both encoding
and decoding, but this patch only includes encoding.
The patch contains further changes by Paweł Wegner
<pawel.wegner95@gmail.com> (primarily for splitting out the encoding
parts of the original patch) and further cleanup, build compatibility
fixes and tweaks for use with Qualcomm encoders by Martin Storsjö.
Signed-off-by: Martin Storsjö <martin@martin.st>
And rename it to retimeinterleave, use the pcm_rechunk bitstream filter for
rechunking.
By seperating the two functions we hopefully get cleaner code.
Signed-off-by: Marton Balint <cus@passwd.hu>
Adding the support to build FFMPEG with HW accelerated decode(nvdec) and
encode on aarch64 architecture.
Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>
The webm_chunk muxer requires the WebM muxer, yet it does not directly
require anything from libavformat/matroska.c (it does not even include
the corresponding header). So remove the dependency from the Makefile
and add a _select to configure.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
vf_dnn_processing.c recently changed to use swscale to trasfer data
between AVFrame and dnn model.
Signed-off-by: Guo, Yejun <yejun.guo@intel.com>
Signed-off-by: Linjie Fu <linjie.fu@intel.com>
Using a compiler with a different host triplet is considered
cross-compiling, even when it is for the same architecture as the
build system. With such a cross-compiler, it is still valid to
optimize builds with --cpu=host. Make the condition that aborts in
this case into a warning instead, since a cross-compiler for an
incompatible architecture will fail with -mtune=native anyway.
Signed-off-by: David Michael <fedora.dm0@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
They use non-public functions, which is unacceptable for a public API
example. Rename the example back to avio_list_dir.
This effectively reverts c84d208c27 and
767d780ec0.
Supports connecting to a RabbitMQ broker via AMQP version 0-9-1.
Signed-off-by: Andriy Gelman <andriy.gelman@gmail.com>
Signed-off-by: Marton Balint <cus@passwd.hu>
The check_x86asm() checks would force enable these variables on success,
bypassing any --disable-* command line option.
This is important in the case of AVX512, where the relevant define is used
to choose between different values for memory alignment and strides in
some allocations.
Signed-off-by: James Almer <jamrial@gmail.com>
This commit adds a chromatic aberration filter for Vulkan that attempts to
emulate a lens chromatic aberration effect.
For a YUV frame it will instead shift the chroma channels, providing a
simple approximation.
This commit adds a Vulkan filtering infrastructure for libavfilter.
It attempts to abstract as much as possible of the Vulkan API from filters.
The way the hwcontext and the framework are designed permits for parallel,
non-CPU-blocking filtering throughout, with the exception of up/downloading
and mapping.
This commit adds the necessary code to initialize and use a Vulkan device
within the hwcontext libavutil framework.
Currently direct mapping to VAAPI and DRM frames is functional, and
transfers to CUDA and native frames are supported.
Lets hope the future Vulkan video decode extension fits well within this
framework.
SetConsoleTextAttribute used to be unavailable for Windows Store apps,
but is available to them now. But GetStdHandle still is unavailable,
thus make sure to check for both functions before using code that
requires both.
Signed-off-by: Martin Storsjö <martin@martin.st>
libx265.c references a member x265_picture.quantOffsets (for ROI
support) which was added in X265_BUILD 70. Increase the minimum libx265
version to fix compilation.
Signed-off-by: Andriy Gelman <andriy.gelman@gmail.com>
Reviewed-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
"VAProcFilterParameterBufferHDRToneMapping" was defined in libva 2.4.1, which will lead to
build failure for the filter tonemap_vaapi for libva 2.3.0 with current check. This patch
is to fix this build error.
Signed-off-by: Xinpeng Sun <xinpeng.sun@intel.com>
When testing on a memory limited system, these tests consume a
significant amount of memory and can often fail if testing by running
multiple processes in parallel.
Signed-off-by: Martin Storsjö <martin@martin.st>
It performs HDR(High Dynamic Range) to SDR(Standard Dynamic Range) conversion
with tone-mapping. It only supports HDR10 as input temporarily.
An example command to use this filter with vaapi codecs:
FFMPEG -hwaccel vaapi -vaapi_device /dev/dri/renderD128 -hwaccel_output_format vaapi \
-i INPUT -vf 'tonemap_vaapi=format=p010' -c:v hevc_vaapi -profile 2 OUTPUT
Signed-off-by: Xinpeng Sun <xinpeng.sun@intel.com>
Signed-off-by: Zachary Zhou <zachary.zhou@intel.com>
Signed-off-by: Ruiling Song <ruiling.song@intel.com>
These functions aren't available when building for the restricted
UWP/WinRT/WinStore API subsets.
Normally when building in this mode, one is probably only building
the libraries, but being able to build ffmpeg.exe still is useful
(and a ffmpeg.exe targeting these API subsets still can be run
e.g. in wine, for testing).
Signed-off-by: Martin Storsjö <martin@martin.st>
fix when pkg-config fail and openssl > 1.1.0 --enable-openssl fail,
the root cause is check_lib can't found the SSL_library_init().
Reviewed-by: James Almer <jamrial@gmail.com>
Signed-off-by: macweng <macweng@tencent.com>
This BSF takes Temporal Units split across different AVPackets and merges them
by looking for Temporal Delimiter OBUs.
Signed-off-by: James Almer <jamrial@gmail.com>
This filter accepts all the dnn networks which do image processing.
Currently, frame with formats rgb24 and bgr24 are supported. Other
formats such as gray and YUV will be supported next. The dnn network
can accept data in float32 or uint8 format. And the dnn network can
change frame size.
The following is a python script to halve the value of the first
channel of the pixel. It demos how to setup and execute dnn model
with python+tensorflow. It also generates .pb file which will be
used by ffmpeg.
import tensorflow as tf
import numpy as np
import imageio
in_img = imageio.imread('in.bmp')
in_img = in_img.astype(np.float32)/255.0
in_data = in_img[np.newaxis, :]
filter_data = np.array([0.5, 0, 0, 0, 1., 0, 0, 0, 1.]).reshape(1,1,3,3).astype(np.float32)
filter = tf.Variable(filter_data)
x = tf.placeholder(tf.float32, shape=[1, None, None, 3], name='dnn_in')
y = tf.nn.conv2d(x, filter, strides=[1, 1, 1, 1], padding='VALID', name='dnn_out')
sess=tf.Session()
sess.run(tf.global_variables_initializer())
output = sess.run(y, feed_dict={x: in_data})
graph_def = tf.graph_util.convert_variables_to_constants(sess, sess.graph_def, ['dnn_out'])
tf.train.write_graph(graph_def, '.', 'halve_first_channel.pb', as_text=False)
output = output * 255.0
output = output.astype(np.uint8)
imageio.imsave("out.bmp", np.squeeze(output))
To do the same thing with ffmpeg:
- generate halve_first_channel.pb with the above script
- generate halve_first_channel.model with tools/python/convert.py
- try with following commands
./ffmpeg -i input.jpg -vf dnn_processing=model=halve_first_channel.model:input=dnn_in:output=dnn_out:fmt=rgb24:dnn_backend=native -y out.native.png
./ffmpeg -i input.jpg -vf dnn_processing=model=halve_first_channel.pb:input=dnn_in:output=dnn_out:fmt=rgb24:dnn_backend=tensorflow -y out.tf.png
Signed-off-by: Guo, Yejun <yejun.guo@intel.com>
Signed-off-by: Pedro Arthur <bygrandao@gmail.com>
1. must enable low_power mode since just VDENC can be supported by iHD
driver right now
2. Coding option1 and extra_data are not supported by MSDK
3. IVF header will be inserted in MSDK by default, but it is not needed
for FFmpeg, so disable it.
Signed-off-by: Zhong Li <zhongli_dev@126.com>
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>
Due to the recent addition of Vulkan support to AMF, we require more
recent headers that include the new structures, which have been
available since AMF 1.4.9 released in September 2018.
Fixes Ticket #8125
Use the command ./configure with/without --disable-v4l2-m2m test.
Reviewed-by: Aman Gupta <aman@tmm1.net>
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>
When compiling natively on an RPI where libomxil-bellagio-dev
was also installed, `check_headers OMX_Core.h` succeeded and
the -isystem compiler flag was never added to the build.
For non-native builds, the error message now mentions the
raspberrypi/firmware repository where the RPI specific
headers are available.
Signed-off-by: Aman Gupta <aman@tmm1.net>
When ffmpeg was streaming, multiple clients were only supported by using a
multicast destination address. An alternative was to stream to a server which
re-distributes the content. This commit adds ZeroMQ as a protocol, which allows
multiple clients to connect to a single ffmpeg instance.
Signed-off-by: Marton Balint <cus@passwd.hu>
The current code in libavfilter/af_sofalizer.c requires
mysofa_neighborhood_init_withstepdefine function, which only appeared
in libmysofa 0.7. Use this function in configure script to bail out
early if a too old libmysofa is found in the system instead of failing
at compile time.
Used a technique similar to lavc/tdsc.c for invoking the MJPEG decoder.
This commit adds support for:
- DNG tiles
- DNG tile huffman lossless JPEG decoding
- DNG 8-bpp ("packed" as dcraw calls it) decoding
- DNG color scaling [1]
- LinearizationTable tag
- BlackLevel tag
[1]: As specified in the DNG Specification - Chapter 5
Signed-off-by: Nick Renieris <velocityra@gmail.com>
Many ffmpeg + rpi compilation guides on the internet recommend
using `./configure --enable-omx --enable-omx-rpi`. This fails
to find the IL OMX headers on device because the omx require_headers
check happens first before the add_cflags in omx_rpi.
A workaround is to use `./configure --enable-omx-rpi` only, since
omx_rpi already implies omx. But because many users expect to use
existing scripts and commands, we swap the order here so omx_rpi
special cases are applied first.
In the past this wasn't an issue because users noticed the OMX_Core.h
missing error and installed libomxil-bellagio-dev. But since
76c82843cc, the rpi specific headers from /opt/vc/include/IL
are required.
Signed-off-by: Aman Gupta <aman@tmm1.net>
MSYS2 converts paths to MinGW-based applications from unix to
pseudo-windows paths on execution time.
Since there was no space between '-include' and the path, MSYS2 doesn't
detect the path properly.
Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>
This avoids using the CUDA SDK at all; instead, we provide a minimal
reimplementation of the basic functionality that lavfi actually uses.
It generates very similar code to what NVCC produces.
The header contains no implementation code derived from the SDK.
The function and type declarations are derived from the SDK only to the
extent required to build a compatible implementation. This is generally
accepted to qualify as fair use.
Because this option does not require the proprietary SDK, it does not require
the "--enable-nonfree" flag in configure.
Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>
Loongson 3A4000 and 2k1000 has supported MSA2.0.
This patch optimized SAD_UB2_UH,UNPCK_R_SH_SW,UNPCK_SB_SH and UNPCK_SH_SW with MSA2.0 instruction.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Get rid of pr dependency and write the columns strictly
alphabetical without page size considerations (POSIX
specifies 66 lines as default).
Setting the page size via pr's -l option was considered,
but as there is issue #5680 which wants to avoid pr
mainly because it's not in busybox, we chose to replace
pr instead.
Before pr would attempt to write pages, thus if a page
boundary was reached, the output looked confusing as one
couldn't see there was a new page and the alphabetical
order was disrupted when scanning down one of the columns.
This change is based on a shell implementation submitted
before by Yejun.
Possible differences to the current version using pr:
1. pr implementations should truncate items to not overflow columns;
depending on how it's done not truncating shall be better IMHO.
2. pr implementations might balance columns differently;
we use minimum number of lines and might end up not
using all columns or might have lesser entries in the
last column(s)
3. we use spaces only for padding the columns; at least the GNU pr
version on my system also by default stuffs in tabs in addition
to a single space in between columns. I don't see that this
behaviour is demanded by POSIX, though I might be very well
overlooking things. Anyway for our use case I can't see a need
for having the additional tabs, or why it would be better compared
to padding with spaces only.
Fixes output for sizes with width < column width, too.
Fixes remaining part of ticket #5680
Contributor: Guo, Yejun <yejun.guo@intel.com>
This patch is based on https://trac.ffmpeg.org/ticket/5680 provided by
Kylie McClain <somasis@exherbo.org> at Wed, 29 Jun 2016 16:37:20 -0400,
and have some changes.
contributor: Kylie McClain <somasis@exherbo.org>
contributor: avih <avihpit@yahoo.com>
Signed-off-by: Guo, Yejun <yejun.guo@intel.com>
Adding the support to build FFMPEG with HW accelerated decode and encode on PPC64
little endian architecture.
Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>
* commit 'c0bd865ad60da31282c5d8e1000c98366249c31e':
configure: Add -D_POSIX_C_SOURCE=200112 -D_XOPEN_SOURCE=600 for mingw as well
Merged-by: James Almer <jamrial@gmail.com>
Mingw headers have got header inline implementations of localtime_r
and gmtime_r, but only visible if certain posix thread safe functions
have been requested.
This is a preparatory step for improving the detection of those
functions.
Signed-off-by: Martin Storsjö <martin@martin.st>