Hybrid 2022.03.20.1: No module named 'vsdpir'

Hybrid 2022.03.20.1: No module named 'vsdpir' - Printable Version

+- Selur's Little Message Board (https://forum.selur.net)
+-- Forum: Hybrid - Support (https://forum.selur.net/forum-1.html)
+--- Forum: Problems & Questions (https://forum.selur.net/forum-3.html)
+--- Thread: Hybrid 2022.03.20.1: No module named 'vsdpir' (/thread-2330.html)

Pages: 1 2 3 4 5 6 7 8

RE: Hybrid 2022.03.20.1: No module named 'vsdpir' - Dan64 - 27.03.2022

I got the same output.

I upgraded vspdir to 2.1.0

This time I got the following error when the preview window is shown:

File "src\cython\vapoursynth.pyx", line 2832, in vapoursynth._vpy_evaluate
File "src\cython\vapoursynth.pyx", line 2833, in vapoursynth._vpy_evaluate
File "F:\VideoTests\Test_vsDpir\tempPreviewVapoursynthFile16_44_23_115.vpy", line 23, in
clip = DPIR(clip=clip, strength=5.000, task="denoise", provider=1, device_id=0)
File "D:\Programs\Hybrid\64bit\Vapoursynth\Lib\site-packages\vsdpir\__init__.py", line 92, in DPIR
raise vs.Error("DPIR: model files have not been downloaded. run 'python -m vsdpir' first")
vapoursynth.Error: DPIR: model files have not been downloaded. run 'python -m vsdpir' first

RE: Hybrid 2022.03.20.1: No module named 'vsdpir' - Selur - 27.03.2022

Okay, seems like the models are missing, call:

python -m vsdpir

Cu Selur

RE: Hybrid 2022.03.20.1: No module named 'vsdpir' - Dan64 - 27.03.2022

(27.03.2022, 16:46)Dan64 Wrote: I got the same output.

I upgraded vspdir to 2.1.0

This time I got the following error when the preview window is shown:

File "src\cython\vapoursynth.pyx", line 2832, in vapoursynth._vpy_evaluate
File "src\cython\vapoursynth.pyx", line 2833, in vapoursynth._vpy_evaluate
File "F:\VideoTests\Test_vsDpir\tempPreviewVapoursynthFile16_44_23_115.vpy", line 23, in
clip = DPIR(clip=clip, strength=5.000, task="denoise", provider=1, device_id=0)
File "D:\Programs\Hybrid\64bit\Vapoursynth\Lib\site-packages\vsdpir\__init__.py", line 92, in DPIR
raise vs.Error("DPIR: model files have not been downloaded. run 'python -m vsdpir' first")
vapoursynth.Error: DPIR: model files have not been downloaded. run 'python -m vsdpir' first

I installed the missing modules with the command "python - m vsdpir first"

But now the preview window is not shown I guess that vsdpir crashed. Maybe there is a problem of compatibility with my card (RTX 3060) but even CPU mode is not working.

I decided to downgrade vsdpir to version 1.6.0 and it is working. I tested all versions (I had to run manually vsViewer.exe) and the last working version is 1.7.1. After that version vsdpir start to crash, I don't know if the problem could be due to the management of TensorRT model (even if this model was already available in version 1.7.1), but you should downgrade vsdipir to version 1.7.1 to improve Hybrid compatibility.

RE: Hybrid 2022.03.20.1: No module named 'vsdpir' - Selur - 27.03.2022

With 2.0 vs-dpir switched to another runtime, which works fine here and you are the first to report problems with it so far, so I will hold off on downgrading to 1.7.1.

The 'Vapoursynth\Lib\site-packages\onnxruntime\capi' folder contains, some dlls from the tensor sdk. My guess is that yes are in the wrong version. (Can't test here since I only own a 1070ti card)

Cu Selur

RE: Hybrid 2022.03.20.1: No module named 'vsdpir' - Dan64 - 27.03.2022

(27.03.2022, 17:27)Selur Wrote: With 2.0 vs-dpir switched to another runtime, which works fine here and you are the first to report problems with it so far, so I will hold off on downgrading to 1.7.1.

Cu Selur

I know, I also tried to unistall and reinstall onnxruntime-gpu, but vs-dpir still crash without issue any error message.

RE: Hybrid 2022.03.20.1: No module named 'vsdpir' - Selur - 27.03.2022

When uninstalling onnxruntime-gpu, did you also remove the 'Vapoursynth\Lib\site-packages\onnxruntime'-folder ?
If not, then you didn't uninstall it completely and the same dlls will be used when reinstalling onnxruntime-gpu .
If you remove that folder you could manually install the CUDA SDK 11.4 (https://developer.nvidia.com/cuda-11-4-4-download-archive only the runtimes need to be installed) and the 'TensorRT 8.0 GA Update 1' (https://developer.nvidia.com/nvidia-tensorrt-8x-download only the runtimes are required)
see: https://onnxruntime.ai/docs/execution-providers/TensorRT-ExecutionProvider.html#requirements

-> if you figure out which dlls are required for TenrsorRT let me know and I can add them to the normal download.

Cu Selur

Ps.: the required dlls can be placed into the 'Vapoursynth\Lib\site-packages\onnxruntime\capi'-folder and the installed runtimes can be deinstalled.

RE: Hybrid 2022.03.20.1: No module named 'vsdpir' - Selur - 27.03.2022

the dlls I need for GTX cards are:

cublas64_11.dll
cublasLt64_11.dll
cudart64_110.dll
cudnn64_8.dll
cudnn_cnn_infer64_8.dll
cudnn_ops_infer64_8.dll
cufft64_10.dll
cufftw64_10.dll

see: https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html
and I thought that

nvinfer.dll
nvinfer_plugin.dll
nvonnxparser.dll
nvparsers.dll

should provide TensorRT support. (Just checked, thise re the only dlls which come with TensorRT-8.0.3.4)

Cu Selur

RE: Hybrid 2022.03.20.1: No module named 'vsdpir' - Dan64 - 27.03.2022

I will try also this solution.

In meanwhile I tried the following steps:

1) Install Hybrid version 2021.12.10.1 with the previous torch archive.
2) Upgraded some of modules to get vs-dpir 2.1.0, my full list of modules if the following

Package            Version

------------------ -----------

addict             2.4.0

bpyutils           0.2.0

certifi            2021.10.8

charset-normalizer 2.0.6

colorama           0.4.4

flatbuffers        2.0

idna               3.2

mmcv-full          1.3.14

numpy              1.22.3

onnxruntime-gpu    1.11.0

opencv-python      4.5.3.56

packaging          21.0

Pillow             8.3.2

pip                22.0.4

protobuf           3.19.4

pyparsing          2.4.7

PyYAML             5.4.1

regex              2021.10.8

requests           2.26.0

setuptools         58.2.0

timm               0.5.4

torch              1.9.1+cu111

torchvision        0.10.1

tqdm               4.62.3

typing-extensions  3.10.0.2

urllib3            1.26.7

VapourSynth        57

vsbasicvsr         1.2.0

vsbasicvsrpp       1.3.0

vsdpir             2.1.0

vsffdnet           1.1.1

vsgan              1.6.4

vshinet            1.0.0

vsrealesrgan       1.2.0

vsrife             1.3.0

vsswinir           1.0.0

vsutil             0.6.0

wheel              0.37.0

yapf               0.31.0

3) Then I installed over the Hybrid version 2022.03.20.1

Now the preview window is working, but it is very slow, it took about 25sec to show the frames. I think that the reason is because I'm using torch 1.9.1+cu111.

But the encoding still crash, I guess that the first frame is provided too late (probably after 25sec). I attached the full debug file.

RE: Hybrid 2022.03.20.1: No module named 'vsdpir' - Selur - 27.03.2022

Speed is slow because CPU rendering is used.
Encoding crashes due to:

x264 output: y4m [error]: bad sequence header magic

x264 [error]: could not open input file `-'

which means, there's a problem with the Vapourynth script, since the script itself:

# Imports

import vapoursynth as vs

# getting Vapoursynth core

core = vs.core

# Loading Plugins

core.std.LoadPlugin(path="D:/Programs/Hybrid/64bit/vsfilters/SourceFilter/LSmashSource/vslsmashsource.dll")

# source: 'F:\VideoTests\Test_vsDpir\Test_video_1m.mkv'

# current color space: YUV420P10, bit depth: 10, resolution: 1280x712, fps: 25, color matrix: 709, yuv luminance scale: limited, scanorder: progressive

# Loading F:\VideoTests\Test_vsDpir\Test_video_1m.mkv using LWLibavSource

clip = core.lsmas.LWLibavSource(source="F:/VideoTests/Test_vsDpir/Test_video_1m.mkv", format="YUV420P10", cache=0, fpsnum=25)

# Setting color matrix to 709.

clip = core.std.SetFrameProps(clip, _Matrix=1)

clip = clip if not core.text.FrameProps(clip,'_Transfer') else core.std.SetFrameProps(clip, _Transfer=1)

clip = clip if not core.text.FrameProps(clip,'_Primaries') else core.std.SetFrameProps(clip, _Primaries=1)

# Setting color range to TV (limited) range.

clip = core.std.SetFrameProp(clip=clip, prop="_ColorRange", intval=1)

# making sure frame rate is set to 25

clip = core.std.AssumeFPS(clip=clip, fpsnum=25, fpsden=1)

from vsdpir import DPIR

# adjusting color space from YUV420P10 to RGBS for vsDPIRDenoise

clip = core.resize.Bicubic(clip=clip, format=vs.RGBS, matrix_in_s="709", range_s="limited")

# denoising using DPIRDenoise

clip = DPIR(clip=clip, strength=5.000, task="denoise", provider=0)

# adjusting output color from: RGBS to YUV420P10 for x264Model

clip = core.resize.Bicubic(clip=clip, format=vs.YUV420P10, matrix_s="709", range_s="limited")

# set output frame rate to 25.000fps

clip = core.std.AssumeFPS(clip=clip, fpsnum=25, fpsden=1)

# Output

clip.set_output()

looks fine to me, the Vapoursynth Preview works (I assume that from your post) and the encoding call:

"D:\Programs\Hybrid\64bit\Vapoursynth\vspipe.exe" "F:\VideoTests\Test_vsDpir\encodingTempSynthSkript_2022-03-27@18_35_18_1610.vpy" - -c y4m | "D:\Programs\Hybrid\64bit\x264.exe" --preset fast --crf 18.00 --profile high10 --level 5.1 --sync-lookahead 30 --psy-rd 1.00:0.15 --vbv-maxrate 240000 --vbv-bufsize 720000 --sar 1:1 --deblock -1:-1 --non-deterministic --range tv --colormatrix bt709 --demuxer y4m --input-range tv --fps 25/1 --opencl --opencl-clbin "F:\VideoTests\Test_vsDpir\x264_lookahead_2022-03-27@18_35_18_1610_02.clbin" --output-depth 10 --output "F:\VideoTests\Test_vsDpir\2022-03-27@18_35_18_1610_02.264" -

my guess is that using the (totally useless) opencl support in x264 interferes with vsdpir and makes it crash somehow.
-> have you tried without the opencl option in x264?

Cu Selur

RE: Hybrid 2022.03.20.1: No module named 'vsdpir' - Dan64 - 27.03.2022

I'm worried that is using the CPU, since I selected "NVIDIA CUDA". The encoding crash even disabling opencl. I tried also x265, same problem.

I will try your solution to add the NVIDIA dlls.