I tried to use NVDECODE API to decode H.265 mpeg-ts video and met some problem. The video is freeze but works well with CPU decoder. I also tried NVDECODE API on H.265 mp4 video, it works well.
I traced the code and found the output from cuvid parser may have some problems to make error.
enter image description here
enter image description here
As you can see, the PicWidthInMbs and FrameHeightInMbs are 0, which is not normal. Any suggestion to resolve it? Or it was a bug from Nvidia?
It is not clear from NVidia CUVID Parser documentation, but i think that this parser doesn't have enough information from ts file.
when i see the second image, it seems that some parsing have been done : BitStreamDataLen, ref_pic_flag, intra_pic_flag, etc... seem to be initialized.
Perhaps, you just need to give to the parser VPS/SPS/PPS data manually, because it seems they are missing with this ts file, or not well parse.
From h265 spec, it seems that the video size comes from SPS data : pic_width_in_luma_samples/pic_height_in_luma_samples.
I will try with h264 ts to get an idea (because i don't have NVidia GPU h265 compliant).
Related
As the title says
I want to know if there are anything that will help someone to trace the laptop or machine used for encoding the video?
Also is there any trace in image file too? Like I watermark with ffmpeg and my machine code is added into metadata of that image?
With ffmpeg, no. Add -bitexact to be sure.
Depending on the applicatzion you are using, container and codec you are encoding to, this is possible.
For ffmpeg i am not aware that it puts any machine related stuff into any format or codec.
Even when you are using external encoders instead of built-in ones like AMD or NVIDIA stuff, currently the codecs do not allow to put such data into the stream.
Sure, future audio/video codecs might allow such metadata in order to find out if the encoder is licensed correctly, but as by now i am not aware of such stuff.
What cameras do for example to overcome the lack of codecs and formats support for storage of this information is to just write some xml along to the media file where they store serial number and such.
If there was such information contained, analyzer tools like "mediainfo" would show this info. I am not yet affiliated with mediainfo Sarl.
Currently, I am trying to use opencv to read a video from my Canon VB-H710F camera.
For this purpose I tried two different solutions:
SOLUTION 1: Read the stream from rtsp address
VideoCapture cam ("rtsp://root:camera#10.0.4.127/stream/profile1=u");
while(true)
cam >> frame;
In this case I am using opencv to directly read from a stream encoded with in H264 (profile1), however this yields the same problem reported here http://answers.opencv.org/question/34012/ip-camera-h264-error-while-decoding/
As suggested in the previous question, I tried to disable FFMPEG support in opencv installation, which solved the h264 decoding errors but raised other problem.
When accessing the stream with opencv, supported by gstreame, there is always a large delay associated.
With this solution I achieve 15 FPS but I have a delay of 5 seconds, which is not acceptable considering that I need a real time application.
SOLUTION 2: Read the frames from http address
while(true)
{
startTime=System.currentTimeMillis();
URL url = new URL("h t t p://[IP]/-wvhttp-01-/image.cgi");
URLConnection con = url.openConnection();
BufferedImage image = ImageIO.read(con.getInputStream());
showImage(image);
estimatedTime=System.currentTimeMillis()-startTime;
System.out.println(estimatedTime);
Thread.sleep(5);
}
This strategy simply grabs the frame from the url that the camera provides. The code is in Java but the results are the same in C++ with the curl library.
This solution avoids the delay of the first solution however it takes little more than 100 ms to grab each frame, which means that I can only achieve on average 10 FPS.
I would like to know how can I read the video using c++ or another library developed in c++ ?
I struggled with similar issues and think I have solved some of your problems using libVLC with OpenCV. FFMPEG seemed to have issues of not decoding H264 properly, plus the newer versions (2.4.11) seemed to have the TCP fix in there already for FFMPEG. Anyways, I use MS Visual Studio on Windows 7 and 8.1.
Details are given here: http://answers.opencv.org/question/65932
Personally, I suggest you to use ffmpeg to read rtsp streams from IP cameras, and then use openCV to read from decoded buffer from ffmpeg. ffmpeg has very good optimizations towards H.264 decoding, performance should not be a critical issue.
You can use ffmpeg binary to verify whether this can work correctly:
ffmpeg -i "rtsp://root:camera#10.0.4.127/stream/profile1=u" -vcodec copy -acodec none test.mp4
If test.mp4 can be played successfully, then it's definitely OK for you to integrate ffmpeg libs into your project.
Good luck!
You can process each frame using ffmpeg as well. you need to create your own filter as per your requirement. https://trac.ffmpeg.org/wiki/FilteringGuide
What I'm trying to do is exactly as the title says, decode multiple compressed audio streams/files - it will be extracted from a modified MP4 file - and do EQ on them in realtime simultaneously.
I have read through most of Apple's docs.
I have tried AudioQueues, but I won't be able to do equalization, as once the compressed audio goes in, it doesn't come out ... so I can't manipulate it.
Audio Units don't seem to have any components to handle decompression of AAC and MP3 - if I'm right it's converter only handles converting from one LPCM format to another.
I have been trying to work out a solution on and off for about a month and a half now.
I'm now thinking, use a 3rd party decoder (god help me; I haven't a clue how to use those, the source code is greek; oh and any recommendations? :x), then feed the decoded-to LPCM into AudioQueues doing EQ at the callback.
Maybe I'm missing something here. Suggestions? :(
I'm still trying to figure out Core Audio for my own needs, but from what I can understand, you want to use Extended Audio File Services which handles reading and compression for you, producing PCM data you can then hand off to a buffer. The MixerHost sample project provides an example of using ExtAudioFileOpenURL to do this.
For latency issues, I would like to send speex encoded audio frame data to a server instead of the raw PCM like I'm sending right now.
The problem is that I'm doing this in flash, and I want to use a socket connection to stream encoded spx frames of data.
I read the speex manual and it unfortunately does not go over the actual CELP algorithm used to convert pcm to spx data, it briefly introduces the use of excitation gains and how it grabs the filter coefficients.
It's libraries are in dlls- dead ends.
I really would like to create a conversion class in actionscript. Is this possible? Is there any documentation on this? I've been googling to no avail. You'd think there would be more documentation on speex out there...
And if I can't do this, what would be the most documente audio format to use?
thanks
I was planning to use the vlc library to decode an H.264 based RTSP stream and extract each frame from it (convert vlc picture to IplImage). I have done a bit of exploration of the vlc code and concluded that there is a function called libvlc_video_take_snapshot which does a similar thing. However the captured frame in this case is saved on the hard disk which I wish to avoid due to the real time nature of my application. What would be the best way to do this? Would it be possible without modifying the vlc source (I want to avoid recompilation if possible). I have heard of vmem etc but could not really figure out what it does and how to use it.
The picture_t structure is internal to the library, how can we get an access to the same.
Awaiting your response.
P.S. Earlier I tried doing this using FFMPEG, however the ffmpeg library has a lot of issues while decoding an H.264 based RTSP stream on windows and hence I had to switch to VLC.
Regards,
Saurabh Gandhi