Migration To Python 3.12/CUDA 12.4 - Success!

[seatv]

I was able to migrate DM without much trouble to Python v3.12.7 and CUDA 12.4. Here are the steps I followed:

Make sure you are on the latest version of pip

py -m pip install --upgrade pip

Install the following in the order below. I removed numpy from dependencies as installing that first caused issues; scikit-image had numpy dependency and it will install numpy.

pip install scikit-image 

pip install opencv_python

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124  

pip install matplotlib 

pip install tensorboardX

The above commands installs the following versions of various packages:

scikit-image - [imageio-2.37.0 lazy-loader-0.4 networkx-3.4.2 numpy-2.2.2 packaging-24.2 pillow-11.1.0 scikit-image-0.25.0 scipy-1.15.1 tifffile-2025.1.10]

opencv_python [4.11.0.86]

pytorch - [MarkupSafe-2.1.5 filelock-3.13.1 fsspec-2024.2.0 jinja2-3.1.3 mpmath-1.3.0 setuptools-70.0.0 sympy-1.13.1 torch-2.5.1+cu124 torchaudio-2.5.1+cu124 torchvision-0.20.1+cu124 typing-extensions-4.9.0]

matplotlib - [contourpy-1.3.1 cycler-0.12.1 fonttools-4.55.3 kiwisolver-1.4.8 matplotlib-3.10.0 pyparsing-3.2.1 python-dateutil-2.9.0.post0 six-1.17.0]

tensorboardX - [ protobuf-5.29.3 tensorboardX-2.6.2.2]

[Tan-Dat-Do]

it's not working for me man. It always caffieGPU missing

[seatv]

Sorry, I was out on vacation. @Wudat123, what GPU do you have?

[Tan-Dat-Do]

My gpu is 3050RTX. So what plugin i need to install?

[seatv]

Sorry for the late response. I think you may have missed installing CUDA Toolkit. You can get it here: https://developer.nvidia.com/cuda-toolkit-archive

This is the explicit link that I used: https://developer.nvidia.com/cuda-12-4-1-download-archive

I assume you have the right driver installed. I use headless NVidia cards, so I just let the above link install the right driver. Some of the cards are do not show up on the driver page.

[seatv]

@Wudat123, please let me know if the above works. There is sample program that the toolkit has that checks to see if you can access the device called deviceQuery. That needs to be built. Other than that, you can run a command line program called

C:\Windows\System32>nvidia-smi
Tue Mar 18 12:54:16 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 560.94                 Driver Version: 560.94         CUDA Version: 12.6     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                  Driver-Model | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce GTX 1060 3GB  WDDM  |   00000000:01:00.0 Off |                  N/A |
| 35%   60C    P2            127W /  120W |    2923MiB /   3072MiB |     31%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A      1956    C+G   ...Brave-Browser\Application\brave.exe      N/A      |
|    0   N/A  N/A      9340    C+G   ...t.LockApp_cw5n1h2txyewy\LockApp.exe      N/A      |
|    0   N/A  N/A     11548    C+G   ...Brave-Browser\Application\brave.exe      N/A      |
|    0   N/A  N/A     11892      C   ...rograms\Python\Python312\python.exe      N/A      |
+-----------------------------------------------------------------------------------------+

In the above, 31% utilization by the python Cuda process I am running.

[HowToDoThis]

Python 3.10.11 + venv + RTX 5070

Note: cause I have installed lots of python version, so I used -V:3.10 specifically
pip should upgrade but my default are v23.0.1

Here are the steps by steps, via command prompt

git clone https://github.com/HypoX64/DeepMosaics.git
cd DeepMosaics

py -V:3.10 -m venv venv
venv\Scripts\activate

pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128
pip install scikit-image opencv_python matplotlib tensorboardX

final output

(venv) F:\Tools\DeepMosaics>python deepmosaic.py --media_path .\mosaicimg.jpg --model_path .\pretrained_models\mosaic\clean_face_HD.pth
makedir: ./result
segment parameters: 12.4M
netG parameters: 182.44M
Clean Mosaic: .\mosaicimg.jpg
Clean temp...
Finished!

(venv) F:\Tools\DeepMosaics>

the installed requirements as the date I written this post

Package           Version
----------------- ------------------------
contourpy         1.3.1
cycler            0.12.1
filelock          3.16.1
fonttools         4.57.0
fsspec            2024.10.0
imageio           2.37.0
Jinja2            3.1.4
kiwisolver        1.4.8
lazy_loader       0.4
MarkupSafe        2.1.5
matplotlib        3.10.1
mpmath            1.3.0
networkx          3.4.2
numpy             2.1.2
opencv-python     4.11.0.86
packaging         24.2
pillow            11.0.0
pip               23.0.1
protobuf          6.30.2
pyparsing         3.2.3
python-dateutil   2.9.0.post0
scikit-image      0.25.2
scipy             1.15.2
setuptools        65.5.0
six               1.17.0
sympy             1.13.3
tensorboardX      2.6.2.2
tifffile          2025.3.30
torch             2.8.0.dev20250404+cu128
torchaudio        2.6.0.dev20250405+cu128
torchvision       0.22.0.dev20250405+cu128
typing_extensions 4.12.2

tq for your steps

[seatv]

@HowToDoThis , Glad to see that you got it working with Cuda 12.8. I have not tried Cuda 12.8!

These days, I have been using lada. It is actively supported, and we get to make feature requests. There is even effort underway to include auto-subtitling (in its infancy). Check this thread out: https://github.com/ladaapp/lada/issues/42

[HowToDoThis]

@HowToDoThis , Glad to see that you got it working with Cuda 12.8. I have not tried Cuda 12.8!

These days, I have been using lada. It is actively supported, and we get to make feature requests. There is even effort underway to include auto-subtitling (in its infancy). Check this thread out: https://github.com/ladaapp/lada/issues/42

I'll try after I get back home, thanks for telling me!

[seatv]

@HowToDoThis - You will need my instructions https://github.com/ladaapp/lada/issues/27 to install on Windows. @ladapp only tests and releases on Linux. If you have high performing hardware, you can install the flatpack Linux version and watch in real-time w/o converting. I have not been able to get lada-gui to work on Windows.

[Tan-Dat-Do]

i use this command pip install -r requirements.txt to install but it seems it require python 3.6 - 3.8

[Tan-Dat-Do]

and also i have install cuda 12.4.0 and it still missing detect_caffe