parsing/decoding video source failed, check GPU memory is available #3770
Comments
|
Can you provide the video source? |
|
resolved |
|
Hello, I met the same problem, how did you solve it? |
There can be many sources to this error, OpenCV version, driver, unsuported video source, unsupported GPU. Can you share your video source? |
The video source is a .mp4 file |
I need the actual file to see if it is supported. Can you also provide
|
Thanks for your response 1 opencv 4.10.0 |
|
I don't have any issues with the following on Windows What is your exact error? |
The code is: camera_path = './src/video/fire/fire1.mp4'
cap = cv2.cudacodec.createVideoReader(camera_path)The specific error output is: /opt/miniconda3/envs/cv/lib/python3.10/site-packages/numpy/core/getlimits.py:549: UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero.
setattr(self, word, getattr(machar, word).flat[0])
/opt/miniconda3/envs/cv/lib/python3.10/site-packages/numpy/core/getlimits.py:89: UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero.
return self._float_to_str(self.smallest_subnormal)
/opt/miniconda3/envs/cv/lib/python3.10/site-packages/numpy/core/getlimits.py:549: UserWarning: The value of the smallest subnormal for <class 'numpy.float32'> type is zero.
setattr(self, word, getattr(machar, word).flat[0])
/opt/miniconda3/envs/cv/lib/python3.10/site-packages/numpy/core/getlimits.py:89: UserWarning: The value of the smallest subnormal for <class 'numpy.float32'> type is zero.
return self._float_to_str(self.smallest_subnormal)
[ERROR:[email protected]] global video_parser.cpp:86 parseVideoData OpenCV(4.10.0-dev) /mnt/e/project/src/opencv/opencv_contrib/modules/cudacodec/src/video_decoder.cpp:124: error: (-217:Gpu API call) CUDA_ERROR_UNKNOWN [Code = 999] in function 'create'
Traceback (most recent call last):
File "/mnt/e/project/code/dev/cloud_platform/cloud-platform/sia-compute-engine/video-analysis-engine/with_ffmpeg/server_test.py", line 14, in <module>
cv2.cudacodec.createVideoReader(camera_path)
cv2.error: OpenCV(4.10.0-dev) /mnt/e/project/src/opencv/opencv_contrib/modules/cudacodec/src/video_reader.cpp:174: error: (-2:Unspecified error) Parsing/Decoding video source failed, check GPU memory is available and GPU supports requested functionality. in function 'waitForDecoderInit' |
|
It looks like your first call to anything which uses the Nvidia decode library fails. This points to an installation or dependancy issue. Can you check that you didn't copy the Nvidia Video Codec SDK stub libraries into a folder on your path? i.e. /usr/local/cuda/lib64 If you did its possible that OpenCV is trying to use a stub .so instead of the one provided by the driver. A quick check would be to look at the output from ldd on your cv2.abi3.so (the exact name may differ depending on how you installed OpenCV)
Notice that libnvcuvid.so.1 => /usr/lib/wsl/lib/libnvcuvid.so.1 (0x00007efdb2d89000) is part of the driver and not located inside /usr/local/cuda/lib64/ like the other CUDA libraries (libnppc.so.12 etc.). |
|
@cudawarped
linux-vdso.so.1 (0x00007ffe4070f000)
libcuda.so.1 => /lib64/libcuda.so.1 (0x00007f8486e34000)
libnvcuvid.so.1 => /lib64/libnvcuvid.so.1 (0x00007f8486200000)
libnvidia-encode.so.1 => /lib64/libnvidia-encode.so.1 (0x00007f8485e00000)
libgstbase-1.0.so.0 => /lib64/libgstbase-1.0.so.0 (0x00007f8485a00000)
libgstreamer-1.0.so.0 => /lib64/libgstreamer-1.0.so.0 (0x00007f8485600000)
libgobject-2.0.so.0 => /lib64/libgobject-2.0.so.0 (0x00007f8485200000)
libglib-2.0.so.0 => /lib64/libglib-2.0.so.0 (0x00007f8484e00000)
libgstapp-1.0.so.0 => /lib64/libgstapp-1.0.so.0 (0x00007f8484a00000)
libgstriff-1.0.so.0 => /lib64/libgstriff-1.0.so.0 (0x00007f8484600000)
libgstpbutils-1.0.so.0 => /lib64/libgstpbutils-1.0.so.0 (0x00007f8484200000)
libgstvideo-1.0.so.0 => /lib64/libgstvideo-1.0.so.0 (0x00007f8483e00000)
libgstaudio-1.0.so.0 => /lib64/libgstaudio-1.0.so.0 (0x00007f8483a00000)
libavcodec.so.58 => /home/itouchtv/soft/ffmpeg/lib/libavcodec.so.58 (0x00007f8482400000)
libavformat.so.58 => /home/itouchtv/soft/ffmpeg/lib/libavformat.so.58 (0x00007f8481e00000)
libavutil.so.56 => /home/itouchtv/soft/ffmpeg/lib/libavutil.so.56 (0x00007f8481800000)
libswscale.so.5 => /home/itouchtv/soft/ffmpeg/lib/libswscale.so.5 (0x00007f8481400000)
libpng16.so.16 => /lib64/libpng16.so.16 (0x00007f8481000000)
libz.so.1 => /lib64/libz.so.1 (0x00007f8480c00000)
libopenblas.so.0 => /lib64/libopenblas.so.0 (0x00007f847e950000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f847e600000)
libnppc.so.11 => /home/itouchtv/cuda/cuda_root/lib64/libnppc.so.11 (0x00007f847e200000)
libnppial.so.11 => /home/itouchtv/cuda/cuda_root/lib64/libnppial.so.11 (0x00007f847d000000)
libnppicc.so.11 => /home/itouchtv/cuda/cuda_root/lib64/libnppicc.so.11 (0x00007f847c600000)
libnppidei.so.11 => /home/itouchtv/cuda/cuda_root/lib64/libnppidei.so.11 (0x00007f847b800000)
libnppif.so.11 => /home/itouchtv/cuda/cuda_root/lib64/libnppif.so.11 (0x00007f8475200000)
libnppig.so.11 => /home/itouchtv/cuda/cuda_root/lib64/libnppig.so.11 (0x00007f8472a00000)
libnppim.so.11 => /home/itouchtv/cuda/cuda_root/lib64/libnppim.so.11 (0x00007f8471e00000)
libnppist.so.11 => /home/itouchtv/cuda/cuda_root/lib64/libnppist.so.11 (0x00007f846f400000)
libnppitc.so.11 => /home/itouchtv/cuda/cuda_root/lib64/libnppitc.so.11 (0x00007f846ec00000)
libcublas.so.11 => /home/itouchtv/cuda/cuda_root/lib64/libcublas.so.11 (0x00007f8468e00000)
libcudnn.so.8 => /home/itouchtv/cuda/cuda_root/lib64/libcudnn.so.8 (0x00007f8468a00000)
libcufft.so.10 => /home/itouchtv/cuda/cuda_root/lib64/libcufft.so.10 (0x00007f8457a00000)
librt.so.1 => /lib64/librt.so.1 (0x00007f8457600000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007f8457200000)
libstdc++.so.6 => /lib64/libstdc++.so.6 (0x00007f8456e00000)
libm.so.6 => /lib64/libm.so.6 (0x00007f8456a00000)
libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007f8456600000)
libc.so.6 => /lib64/libc.so.6 (0x00007f8456200000)
/lib64/ld-linux-x86-64.so.2 (0x00007f8495ba5000)
libgmodule-2.0.so.0 => /lib64/libgmodule-2.0.so.0 (0x00007f8455e00000)
libffi.so.6 => /lib64/libffi.so.6 (0x00007f8455a00000)
libpcre.so.1 => /lib64/libpcre.so.1 (0x00007f8455600000)
libgsttag-1.0.so.0 => /lib64/libgsttag-1.0.so.0 (0x00007f8455200000)
liborc-0.4.so.0 => /lib64/liborc-0.4.so.0 (0x00007f8454e00000)
libswresample.so.3 => /home/itouchtv/soft/ffmpeg/lib/libswresample.so.3 (0x00007f8454a00000)
liblzma.so.5 => /lib64/liblzma.so.5 (0x00007f8454600000)
libbz2.so.1 => /lib64/libbz2.so.1 (0x00007f8454200000)
libgfortran.so.5 => /lib64/libgfortran.so.5 (0x00007f8453c00000)
libcublasLt.so.11 => /home/itouchtv/cuda/cuda_root/lib64/libcublasLt.so.11 (0x00007f842f600000)
libquadmath.so.0 => /lib64/libquadmath.so.0 (0x00007f842f200000) |
|
Did you copy /lib64/libnvidia-encode.so.1 and /lib64/libnvcuvid.so.1 from the Lib/linux/stubs directory or are they part of the driver? If you copied them then they need to be removed as they are stubs provided for building on systems without the Nvidia driver installed. i.e. they don't contain any implementation code for accessing the hardware encoder/decoder. |
No, I did not copy them from Video_Codec_SDK_12.2.72/Lib/linux/stubs/x86_64 to /lib64, they existed likely with the driver installed together, as following: lrwxrwxrwx 1 root root 21 Dec 5 17:42 /lib64/libnvcuvid.so.1 -> libnvcuvid.so.550.135
lrwxrwxrwx 1 root root 27 Dec 5 17:42 /lib64/libnvidia-encode.so.1 -> libnvidia-encode.so.550.135But the other thing which I have done was that I surely copied the .so files from Video_Codec_SDK_12.2.72/Lib/linux/stubs/x86_64 to <CUDA_PATH>/lib64/stubs, like this: -rw-rw-r-- 1 itouchtv itouchtv 3528 Sep 9 15:58 /home/itouchtv/cuda/cuda_root/lib64/stubs/libnvcuvid.so
-rw-rw-r-- 1 itouchtv itouchtv 1480 Sep 9 15:58 /home/itouchtv/cuda/cuda_root/lib64/stubs/libnvidia-encode.so
So what is the next step for me, do I need to try to remove the symbolic links? |
Copying the stubs to the stubs directory should be fine as you can see from the output in ldd your python bindings should not be using them. That said to be 100% you could just remove them and see if it runs in case for some reason they are being added to a path at runtime by micromamba. Are you getting the exact same error as described above? |
|
Actually my first try is to rename the cuda_root/lib64/stubs/libnvcuvid.so to another one, but does not work. All the content of error I encountered is shown below: $ ipython
Python 3.10.14 | packaged by conda-forge | (main, Mar 20 2024, 12:45:18) [GCC 12.3.0]
Type 'copyright', 'credits' or 'license' for more information
IPython 8.27.0 -- An enhanced Interactive Python. Type '?' for help.
In [1]: import cv2
In [2]: video_path = '~/hpt/output.mp4'
In [3]: video_capture_cuda = cv2.cudacodec.createVideoReader(video_path)
[ERROR:0@39.220] global video_parser.cpp:86 parseVideoData OpenCV(4.10.0) /home/itouchtv/hpt/opencv_src/opencv-python/opencv_contrib/modules/cudacodec/src/video_decoder.cpp:149: error: (-215:Assertion failed) videoFormat.ulWidth >= decodeCaps.nMinWidth && videoFormat.ulHeight >= decodeCaps.nMinHeight && videoFormat.ulWidth <= decodeCaps.nMaxWidth && videoFormat.ulHeight <= decodeCaps.nMaxHeight in function 'create'
---------------------------------------------------------------------------
error Traceback (most recent call last)
Cell In[3], line 1
----> 1 video_capture_cuda = cv2.cudacodec.createVideoReader(video_path)
error: OpenCV(4.10.0) /home/itouchtv/hpt/opencv_src/opencv-python/opencv_contrib/modules/cudacodec/src/video_reader.cpp:174: error: (-2:Unspecified error) Parsing/Decoding video source failed, check GPU memory is available and GPU supports requested functionality. in function 'waitForDecoderInit'
In [4]: video_capture = cv2.VideoCapture(video_path)
In [6]: video_capture.get(cv2.CAP_PROP_FRAME_WIDTH)
Out[6]: 4724.0
In [7]: video_capture.get(cv2.CAP_PROP_FRAME_HEIGHT)
Out[7]: 2656.0
In [8]: _, frame = video_capture.read()
In [10]: cv2.imwrite('frame.jpg', frame)
Out[10]: TrueIn other words, when the video resolution exceeds 3840 * 2160, my OpenCV using GPU for frame extraction encounters this error, while using the CPU works normally; if the video resolution is equal to or less than 3840 * 2160, both work normally. However, generally speaking, when the video resolution reaches above 3840 * 2160, GPU processing speed is much faster than CPU, so I want to resolve this issue. |
|
The error is not the same but is explicit in the cause
The message could be a little more explicit but its saying that your hardware decoder does not support the width or height of the encoded video. Refer to the Video Decoder Capabilities table for a list of supported resolutions by codec. What codec are you using? If the table indicates that the video should be supported can you share the video so I can test it? |
|
@cudawarped
lspci | grep -i NVIDIAOutput 1: 00:08.0 3D controller: NVIDIA Corporation GA102GL [A10] (rev a1)Output 2: 01:00.0 VGA compatible controller: NVIDIA Corporation TU104GL [Quadro RTX 4000] (rev a1)
ffmpeg -i output.mp4Output : Stream #0:0(eng): Video: h264 (Main) (avc1 / 0x31637661), yuv420p(tv, bt709), 4724x2656, 9998 kb/s, 25 fps, 25 tbr, 25k tbn, 50 tbc (default)Comparing the table requirements, my GPU architecture is in the last two rows, find the column corresponding to h264 (column name is H.264/AVCHD)
The result is: Maximum Resolution: 4096 x 4096 So the conclusion should be: The maximum resolution of the H264 codec video supported by my GPU architecture is 4096 x 4096, and a width or height greater than this value will not work. Thank you very much for your help and have fun. |
System information (version)
Detailed description
error when decode video exceeding 1920*1080.
video_reader.cpp:174:error:(-2:Unspecified error) Parsing/Decoding video source failed, check GPU memory is available and GPU supports requested functionality. in function 'waitForDecoderInit'
Steps to reproduce
The text was updated successfully, but these errors were encountered: