Torch == 2.5.1 with Cuda + torchvision can't be used with opencv

[lordsoffallen]

🐛 Describe the bug

torchvision 0.20.0 and 0.20.1 has a requirement of ffmpeg to be >=4.4.2, <5.0.0 which is not compatible with opencv as opencv requires higher versions which creates environment conflict and in turn results in downgrading pytorch to 2.4.0 and torchvision to 0.19.0.

Here is a full preview of installation plan:

Channels:
 - nvidia
 - pytorch
 - conda-forge
 - defaults
Platform: linux-64
Collecting package metadata (repodata.json): done
Solving environment: done

## Package Plan ##

  environment location: /home/ftopal/miniconda3/envs/projx-gpu

  added / updated specs:
    - opencv


The following packages will be downloaded:

    package                    |            build
    ---------------------------|-----------------
    ffmpeg-7.1.0               | gpl_h4c12d27_707         9.9 MB  conda-forge
    gdk-pixbuf-2.42.12         |       hb9ae30d_0         516 KB  conda-forge
    hdf5-1.14.4                |nompi_h2d575fe_105         3.8 MB  conda-forge
    libgfortran-14.2.0         |       h69a702a_1          53 KB  conda-forge
    libgfortran5-14.2.0        |       hd5240d6_1         1.4 MB  conda-forge
    libglu-9.0.3               |       h03adeef_0         318 KB  conda-forge
    libhwloc-2.11.2            |default_h0d58e46_1001         2.3 MB  conda-forge
    libopencv-4.10.0           |qt6_py310h0bc352f_613        29.5 MB  conda-forge
    libopenvino-2024.5.0       |       hac27bb2_0         5.3 MB  conda-forge
    libopenvino-auto-batch-plugin-2024.5.0|       h4d9b6c2_0         108 KB  conda-forge
    libopenvino-auto-plugin-2024.5.0|       h4d9b6c2_0         232 KB  conda-forge
    libopenvino-hetero-plugin-2024.5.0|       h3f63f65_0         192 KB  conda-forge
    libopenvino-intel-cpu-plugin-2024.5.0|       hac27bb2_0        11.7 MB  conda-forge
    libopenvino-intel-gpu-plugin-2024.5.0|       hac27bb2_0         9.1 MB  conda-forge
    libopenvino-intel-npu-plugin-2024.5.0|       hac27bb2_0         946 KB  conda-forge
    libopenvino-ir-frontend-2024.5.0|       h3f63f65_0         203 KB  conda-forge
    libopenvino-onnx-frontend-2024.5.0|       h5c8f2c3_0         1.5 MB  conda-forge
    libopenvino-paddle-frontend-2024.5.0|       h5c8f2c3_0         643 KB  conda-forge
    libopenvino-pytorch-frontend-2024.5.0|       h5888daf_0         1.1 MB  conda-forge
    libopenvino-tensorflow-frontend-2024.5.0|       h6481b9d_0         1.3 MB  conda-forge
    libopenvino-tensorflow-lite-frontend-2024.5.0|       h5888daf_0         474 KB  conda-forge
    librsvg-2.58.4             |       hc0ffecb_0         6.1 MB  conda-forge
    ocl-icd-2.3.2              |       hb9d3cd8_2          93 KB  conda-forge
    opencl-headers-2024.10.24  |       h5888daf_0          53 KB  conda-forge
    opencv-4.10.0              |qt6_py310ha05f5c2_613          26 KB  conda-forge
    openexr-3.3.2              |       h6326327_1         1.3 MB  conda-forge
    openh264-2.5.0             |       hf92e6e3_0         710 KB  conda-forge
    pango-1.54.0               |       h3a902e7_3         437 KB  conda-forge
    py-opencv-4.10.0           |qt6_py310h4bc6995_613         1.1 MB  conda-forge
    pyside6-6.7.3              |  py310hfd10a26_1        10.0 MB  conda-forge
    pytorch-2.4.0              |py3.10_cuda12.4_cudnn9.1.0_0        1.34 GB  pytorch
    qt6-main-6.7.3             |       h6e8976b_1        45.2 MB  conda-forge
    svt-av1-2.3.0              |       h5888daf_0         2.6 MB  conda-forge
    tbb-2022.0.0               |       hceb3a55_0         174 KB  conda-forge
    torchtriton-3.0.0          |            py310       233.4 MB  pytorch
    torchvision-0.19.0         |      py310_cu124         8.2 MB  pytorch
    ------------------------------------------------------------
                                           Total:        1.72 GB

The following NEW packages will be INSTALLED:

  dav1d              conda-forge/linux-64::dav1d-1.2.1-hd590300_0 
  freeglut           conda-forge/linux-64::freeglut-3.2.2-ha6d2627_3 
  fribidi            conda-forge/linux-64::fribidi-1.0.10-h36c2ea0_0 
  gdk-pixbuf         conda-forge/linux-64::gdk-pixbuf-2.42.12-hb9ae30d_0 
  hdf5               conda-forge/linux-64::hdf5-1.14.4-nompi_h2d575fe_105 
  imath              conda-forge/linux-64::imath-3.1.12-h7955e40_0 
  jasper             conda-forge/linux-64::jasper-4.2.4-h536e39c_0 
  libaec             conda-forge/linux-64::libaec-1.1.3-h59595ed_0 
  libass             conda-forge/linux-64::libass-0.17.3-h1dc1e6a_0 
  libgfortran        conda-forge/linux-64::libgfortran-14.2.0-h69a702a_1 
  libgfortran5       conda-forge/linux-64::libgfortran5-14.2.0-hd5240d6_1 
  libglu             conda-forge/linux-64::libglu-9.0.3-h03adeef_0 
  libhwloc           conda-forge/linux-64::libhwloc-2.11.2-default_h0d58e46_1001 
  liblapacke         conda-forge/linux-64::liblapacke-3.9.0-16_linux64_mkl 
  libopencv          conda-forge/linux-64::libopencv-4.10.0-qt6_py310h0bc352f_613 
  libopenvino        conda-forge/linux-64::libopenvino-2024.5.0-hac27bb2_0 
  libopenvino-auto-~ conda-forge/linux-64::libopenvino-auto-batch-plugin-2024.5.0-h4d9b6c2_0 
  libopenvino-auto-~ conda-forge/linux-64::libopenvino-auto-plugin-2024.5.0-h4d9b6c2_0 
  libopenvino-heter~ conda-forge/linux-64::libopenvino-hetero-plugin-2024.5.0-h3f63f65_0 
  libopenvino-intel~ conda-forge/linux-64::libopenvino-intel-cpu-plugin-2024.5.0-hac27bb2_0 
  libopenvino-intel~ conda-forge/linux-64::libopenvino-intel-gpu-plugin-2024.5.0-hac27bb2_0 
  libopenvino-intel~ conda-forge/linux-64::libopenvino-intel-npu-plugin-2024.5.0-hac27bb2_0 
  libopenvino-ir-fr~ conda-forge/linux-64::libopenvino-ir-frontend-2024.5.0-h3f63f65_0 
  libopenvino-onnx-~ conda-forge/linux-64::libopenvino-onnx-frontend-2024.5.0-h5c8f2c3_0 
  libopenvino-paddl~ conda-forge/linux-64::libopenvino-paddle-frontend-2024.5.0-h5c8f2c3_0 
  libopenvino-pytor~ conda-forge/linux-64::libopenvino-pytorch-frontend-2024.5.0-h5888daf_0 
  libopenvino-tenso~ conda-forge/linux-64::libopenvino-tensorflow-frontend-2024.5.0-h6481b9d_0 
  libopenvino-tenso~ conda-forge/linux-64::libopenvino-tensorflow-lite-frontend-2024.5.0-h5888daf_0 
  libopus            conda-forge/linux-64::libopus-1.3.1-h7f98852_1 
  librsvg            conda-forge/linux-64::librsvg-2.58.4-hc0ffecb_0 
  ocl-icd            conda-forge/linux-64::ocl-icd-2.3.2-hb9d3cd8_2 
  opencl-headers     conda-forge/linux-64::opencl-headers-2024.10.24-h5888daf_0 
  opencv             conda-forge/linux-64::opencv-4.10.0-qt6_py310ha05f5c2_613 
  openexr            conda-forge/linux-64::openexr-3.3.2-h6326327_1 
  pango              conda-forge/linux-64::pango-1.54.0-h3a902e7_3 
  pugixml            conda-forge/linux-64::pugixml-1.14-h59595ed_0 
  py-opencv          conda-forge/linux-64::py-opencv-4.10.0-qt6_py310h4bc6995_613 
  tbb                conda-forge/linux-64::tbb-2022.0.0-hceb3a55_0 

The following packages will be UPDATED:

  aom                                      3.6.1-h59595ed_0 --> 3.9.1-hac33072_0 
  ffmpeg                             4.4.2-gpl_hdf48244_113 --> 7.1.0-gpl_h4c12d27_707 
  libvpx                                  1.13.1-h59595ed_0 --> 1.14.1-hac33072_0 
  openh264                                 2.3.1-hcb278e6_2 --> 2.5.0-hf92e6e3_0 
  svt-av1                                  1.4.1-hcb278e6_0 --> 2.3.0-h5888daf_0 

The following packages will be DOWNGRADED:

  pyside6                             6.8.1-py310hfd10a26_0 --> 6.7.3-py310hfd10a26_1 
  pytorch                2.5.1-py3.10_cuda12.4_cudnn9.1.0_0 --> 2.4.0-py3.10_cuda12.4_cudnn9.1.0_0 
  qt6-main                                 6.8.1-h9d28a51_0 --> 6.7.3-h6e8976b_1 
  torchtriton                                   3.1.0-py310 --> 3.0.0-py310 
  torchvision                            0.20.1-py310_cu124 --> 0.19.0-py310_cu124 

Versions

torchvision=0.20.1-py310_cu124
pytorch=2.5.1-py3.10_cuda12.4_cudnn9.1.0_0