Previously, the demuxer would register decoder with the scheduler, using
InputStream as opaque, and pass the scheduling index to the decoder.
Now the registration is done by the decoder itself, using DecoderPriv as
opaque, and the scheduling index is returned to demuxer from dec_open().
decoder_thread() then no longer needs to be accessed from outside of
ffmpeg_dec and can be made static.
It is done based on demuxer information, so that is the more appropriate
place for this code.
This is a step towards decoupling Decoder and InputStream.
* as this decision is based on demuxing information, move it from the
decoder to the demuxer
* as the issue being addressed is latency added by frame threading, we
only need to disable frame threading, not all threading
Similar to what is currently done for other components, e.g. (de)muxers.
There is nothing in the public part currently, but that will change in
future commits.
The old code did not follow the syntax from the spec.
Reviewed-by: Frank Plowman <post@frankplowman.com>
Signed-off-by: James Almer <jamrial@gmail.com>
Use 8 packets/frames by default rather than 1, which seems to provide
better throughput.
Allow -thread_queue_size to set the muxer queue size manually again.
It presumably exists because HapContext contains an AVClass*.
Yet AVClass is actually defined in log.h and even this inclusion
can be avoided by struct AVClass*. This avoids opt.h inclusions
in hap.c and hapdec.c.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
ff_texturedspenc_init() doesn't support most of the function types
supported for decoding; add a separate context containing only pointers
for the actually supported types.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The old code set a common struct from each thread;
this only "worked" (but is still UB) because the values
written are the same for each thread.
Fix this by moving the assignments to the main thread.
(This also avoids casting const away from a const AVFrame*.)
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Use class confidence instead of box_score to filt boxes, which is more
accurate. Class confidence is obtained by multiplying class probability
distribution and box_score.
Signed-off-by: Wenbin Chen <wenbin.chen@intel.com>
Reviewed-by: Guo Yejun <yejun.guo@intel.com>
For detect and classify output, width and height make no sence, so
change width, height to dims to represent the shape of tensor. Use
layout and dims to get width, height and channel.
Signed-off-by: Wenbin Chen <wenbin.chen@intel.com>
Reviewed-by: Guo Yejun <yejun.guo@intel.com>
Now when using openvino backend, user doesn't need to set input/output
names in command line. Model ports will be automatically detected.
For example:
ffmpeg -i input.png -vf \
dnn_detect=dnn_backend=openvino:model=model.xml:input=image:\
output=detection_out -y output.png
can be simplified to:
ffmpeg -i input.png -vf dnn_detect=dnn_backend=openvino:model=model.xml\
-y output.png
Signed-off-by: Wenbin Chen <wenbin.chen@intel.com>
Reviewed-by: Guo Yejun <yejun.guo@intel.com>
nal->skipped_bytes_pos contains the positions of errors relative to the
start of the slice header, whereas the position they were tested against
is relative to the start of the slice data, i.e. one byte after the end
of the slice header.
Patch fixes this by storing the size of the slice header in H266RawSlice
and adding it to the position given by the GetBitContext before
comparing to skipped_bytes_pos. This fixes AVERROR_INVALIDDATAs for
various valid bitstreams, such as the LMCS_B_Dolby_2 conformance
bitstream.
Signed-off-by: Frank Plowman <post@frankplowman.com>
Resolves the following undefined behavior sanitiser error:
runtime error: shift exponent 32 is too large for 32-bit type 'int'
Signed-off-by: Frank Plowman <post@frankplowman.com>
Signed-off-by: James Almer <jamrial@gmail.com>
Anton Khirnov
James Almer
Frank Plowman
Andreas Rheinhardt
Wenbin Chen