Sanster/IOPaint

[BUG] GFPGAN error when GFPGAN is installed

SoftologyPro opened this issue · 12 comments

SoftologyPro commented

Using this command

iopaint start --model=lama --device=cuda --port=8080 --enable-interactive-seg --interactive-seg-model=vit_l --interactive-seg-device=cuda --enable-anime-seg --enable-gfpgan --gfpgan-device cuda --enable-remove-bg

gives this error

2024-03-04 09:45:19.135 | INFO     | iopaint.runtime:setup_model_dir:82 - Model directory: C:\Users\Jason\.cache
- Platform: Windows-10-10.0.22621-SP0
- Python version: 3.10.8
- torch: 2.2.0+cu118
- torchvision: 0.17.0+cu118
- Pillow: 9.5.0
- diffusers: 0.26.3
- transformers: 4.38.2
- opencv-python: 4.9.0.80
- accelerate: 0.27.2
- iopaint: 1.2.1
- rembg: 2.0.55
- realesrgan: 0.3.0
- gfpgan: 1.3.8

[W init.cpp:768] Warning: nvfuser is no longer supported in torch script, use _jit_set_nvfuser_enabled is deprecated and a no-op (function operator ())
2024-03-04 09:45:21.097 | INFO     | iopaint.plugins:build_plugins:32 - Initialize InteractiveSeg plugin
2024-03-04 09:45:21.097 | INFO     | iopaint.plugins.interactive_seg:_init_session:62 - SegmentAnything model path: C:\Users\Jason\.cache\torch\hub\checkpoints\sam_vit_l_0b3195.pth
2024-03-04 09:45:22.858 | INFO     | iopaint.plugins:build_plugins:38 - Initialize RemoveBG plugin
2024-03-04 09:45:23.902 | INFO     | iopaint.plugins:build_plugins:42 - Initialize AnimeSeg plugin
2024-03-04 09:45:24.008 | INFO     | iopaint.helper:load_model:123 - Loading model from: C:\Users\Jason\.cache\torch\hub\checkpoints\isnetis.pth
2024-03-04 09:45:24.104 | INFO     | iopaint.plugins:build_plugins:56 - Initialize GFPGAN plugin
2024-03-04 09:45:24.104 | INFO     | iopaint.plugins:build_plugins:60 - GFPGAN no background upscaler, use --enable-realesrgan to enable it
2024-03-04 09:45:24.281 | ERROR    | iopaint.plugins.base_plugin:__init__:15 - gfpgan is not installed, please install it first. pip install gfpgan
{
    "host": "127.0.0.1",
    "port": 8080,
    "inbrowser": false,
    "model": "lama",
    "no_half": false,
    "low_mem": false,
    "cpu_offload": false,
    "disable_nsfw_checker": false,
    "local_files_only": false,
    "cpu_textencoder": false,
    "device": "cuda",
    "input": null,
    "output_dir": null,
    "quality": 95,
    "enable_interactive_seg": true,
    "interactive_seg_model": "vit_l",
    "interactive_seg_device": "cuda",
    "enable_remove_bg": true,
    "remove_bg_model": "briaai/RMBG-1.4",
    "enable_anime_seg": true,
    "enable_realesrgan": false,
    "realesrgan_device": "cpu",
    "realesrgan_model": "realesr-general-x4v3",
    "enable_gfpgan": true,
    "gfpgan_device": "cuda",
    "enable_restoreformer": false,
    "restoreformer_device": "cpu"
}

You can see gfpgan is installed as it shows it is version 1.3.8.
GFPGAN was installed with pip install gfpgan

What am I missing?

SoftologyPro commented

Similar error for realesrgan. I lauinch with this command

iopaint start --model=lama --device=cuda --port=8080 --enable-interactive-seg --interactive-seg-model=vit_l --interactive-seg-device=cuda --enable-anime-seg --enable-realesrgan --realesrgan-model RealESRGAN_x4plus --realesrgan-device cuda --enable-remove-bg

and get this error (you can see realesrgan is detected as installed)

2024-03-04 09:49:37.458 | INFO     | iopaint.runtime:setup_model_dir:82 - Model directory: C:\Users\Jason\.cache
- Platform: Windows-10-10.0.22621-SP0
- Python version: 3.10.8
- torch: 2.2.0+cu118
- torchvision: 0.17.0+cu118
- Pillow: 9.5.0
- diffusers: 0.26.3
- transformers: 4.38.2
- opencv-python: 4.9.0.80
- accelerate: 0.27.2
- iopaint: 1.2.1
- rembg: 2.0.55
- realesrgan: 0.3.0
- gfpgan: 1.3.8

[W init.cpp:768] Warning: nvfuser is no longer supported in torch script, use _jit_set_nvfuser_enabled is deprecated and a no-op (function operator ())
2024-03-04 09:49:39.317 | INFO     | iopaint.plugins:build_plugins:32 - Initialize InteractiveSeg plugin
2024-03-04 09:49:39.317 | INFO     | iopaint.plugins.interactive_seg:_init_session:62 - SegmentAnything model path: C:\Users\Jason\.cache\torch\hub\checkpoints\sam_vit_l_0b3195.pth
2024-03-04 09:49:41.083 | INFO     | iopaint.plugins:build_plugins:38 - Initialize RemoveBG plugin
2024-03-04 09:49:42.154 | INFO     | iopaint.plugins:build_plugins:42 - Initialize AnimeSeg plugin
2024-03-04 09:49:42.261 | INFO     | iopaint.helper:load_model:123 - Loading model from: C:\Users\Jason\.cache\torch\hub\checkpoints\isnetis.pth
2024-03-04 09:49:42.359 | INFO     | iopaint.plugins:build_plugins:46 - Initialize RealESRGAN plugin: RealESRGAN_x4plus, cuda
2024-03-04 09:49:42.527 | ERROR    | iopaint.plugins.base_plugin:__init__:15 - RealESRGAN is not installed, please install it first. pip install realesrgan
{
    "host": "127.0.0.1",
    "port": 8080,
    "inbrowser": false,
    "model": "lama",
    "no_half": false,
    "low_mem": false,
    "cpu_offload": false,
    "disable_nsfw_checker": false,
    "local_files_only": false,
    "cpu_textencoder": false,
    "device": "cuda",
    "input": null,
    "output_dir": null,
    "quality": 95,
    "enable_interactive_seg": true,
    "interactive_seg_model": "vit_l",
    "interactive_seg_device": "cuda",
    "enable_remove_bg": true,
    "remove_bg_model": "briaai/RMBG-1.4",
    "enable_anime_seg": true,
    "enable_realesrgan": true,
    "realesrgan_device": "cuda",
    "realesrgan_model": "RealESRGAN_x4plus",
    "enable_gfpgan": false,
    "gfpgan_device": "cpu",
    "enable_restoreformer": false,
    "restoreformer_device": "cpu"
}
  • 👍1
SoftologyPro commented

Also same if I try restoreformer (but this complaiins about GFPGAN again)

iopaint start --model=lama --device=cuda --port=8080 --enable-interactive-seg --interactive-seg-model=vit_l --interactive-seg-device=cuda --enable-anime-seg --enable-remove-bg --enable-restoreformer --restoreformer-device cuda

error is

2024-03-04 09:51:55.078 | INFO     | iopaint.runtime:setup_model_dir:82 - Model directory: C:\Users\Jason\.cache
- Platform: Windows-10-10.0.22621-SP0
- Python version: 3.10.8
- torch: 2.2.0+cu118
- torchvision: 0.17.0+cu118
- Pillow: 9.5.0
- diffusers: 0.26.3
- transformers: 4.38.2
- opencv-python: 4.9.0.80
- accelerate: 0.27.2
- iopaint: 1.2.1
- rembg: 2.0.55
- realesrgan: 0.3.0
- gfpgan: 1.3.8

[W init.cpp:768] Warning: nvfuser is no longer supported in torch script, use _jit_set_nvfuser_enabled is deprecated and a no-op (function operator ())
2024-03-04 09:51:56.956 | INFO     | iopaint.plugins:build_plugins:32 - Initialize InteractiveSeg plugin
2024-03-04 09:51:56.956 | INFO     | iopaint.plugins.interactive_seg:_init_session:62 - SegmentAnything model path: C:\Users\Jason\.cache\torch\hub\checkpoints\sam_vit_l_0b3195.pth
2024-03-04 09:51:58.679 | INFO     | iopaint.plugins:build_plugins:38 - Initialize RemoveBG plugin
2024-03-04 09:52:00.270 | INFO     | iopaint.plugins:build_plugins:42 - Initialize AnimeSeg plugin
2024-03-04 09:52:00.386 | INFO     | iopaint.helper:load_model:123 - Loading model from: C:\Users\Jason\.cache\torch\hub\checkpoints\isnetis.pth
2024-03-04 09:52:00.489 | INFO     | iopaint.plugins:build_plugins:69 - Initialize RestoreFormer plugin
2024-03-04 09:52:00.662 | ERROR    | iopaint.plugins.base_plugin:__init__:15 - gfpgan is not installed, please install it first. pip install gfpgan
{
    "host": "127.0.0.1",
    "port": 8080,
    "inbrowser": false,
    "model": "lama",
    "no_half": false,
    "low_mem": false,
    "cpu_offload": false,
    "disable_nsfw_checker": false,
    "local_files_only": false,
    "cpu_textencoder": false,
    "device": "cuda",
    "input": null,
    "output_dir": null,
    "quality": 95,
    "enable_interactive_seg": true,
    "interactive_seg_model": "vit_l",
    "interactive_seg_device": "cuda",
    "enable_remove_bg": true,
    "remove_bg_model": "briaai/RMBG-1.4",
    "enable_anime_seg": true,
    "enable_realesrgan": false,
    "realesrgan_device": "cpu",
    "realesrgan_model": "realesr-general-x4v3",
    "enable_gfpgan": false,
    "gfpgan_device": "cpu",
    "enable_restoreformer": true,
    "restoreformer_device": "cuda"
}
SoftologyPro commented

If I remove those 3 parameters and launch with

iopaint start --model=lama --device=cuda --port=8080 --enable-interactive-seg --interactive-seg-model=vit_l --interactive-seg-device=cuda --enable-anime-seg --enable-remove-bg

It starts and runs fine.

Sanster commented

The logs of iopaint did not print the complete error message. Can you try manually importing import gfpgan and import realesrgan to see what error occurs?

SoftologyPro commented

The logs of iopaint did not print the complete error message. Can you try manually importing import gfpgan and import realesrgan to see what error occurs?

How do I add those inport lines? I setup IOPaint with pip install iopaint, so there is no code to modify.

Specificaly, I setup a new clean environment and then install the requirements with

python -m pip install --upgrade pip
pip install --no-cache-dir --ignore-installed --force-reinstall --no-warn-conflicts wheel==0.42.0
pip install --no-cache-dir --ignore-installed --force-reinstall --no-warn-conflicts rembg==2.0.55
pip install --no-cache-dir --ignore-installed --force-reinstall --no-warn-conflicts realesrgan==0.3.0
pip install --no-cache-dir --ignore-installed --force-reinstall --no-warn-conflicts gfpgan==1.3.8
pip install --no-cache-dir --ignore-installed --force-reinstall --no-warn-conflicts iopaint==1.2.1
pip uninstall -y torch
pip install --no-cache-dir --ignore-installed --force-reinstall --no-warn-conflicts torch==2.2.0+cu118 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip uninstall -y charset-normalizer
pip install --no-cache-dir --ignore-installed --force-reinstall --no-warn-conflicts charset-normalizer==3.3.2
pip uninstall -y typing_extensions
pip install --no-cache-dir --ignore-installed --force-reinstall --no-warn-conflicts typing_extensions==4.8.0
Sanster commented

Create a python file, name it whatever you want, for example tmp.py, with only one line of content:

import gfpgan

Then execute python tmp.py, this command should produce an error, please paste the error message.

SoftologyPro commented

ModuleNotFoundError: No module named 'torchvision.transforms.functional_tensor'

Does it need a torch version older than 2.2.0?

OK, fixed, it needs Torch 2.1.0 to work. Maybe specify the exact version in your requirements.
https://softologyblog.wordpress.com/2023/10/10/a-plea-to-all-python-developers/

RealESRGAN and RestoreFormer work fine now too.

Thanks for your help.

Sanster commented

ModuleNotFoundError: No module named 'torchvision.transforms.functional_tensor'

Does it need a torch version older than 2.2.0?

OK, fixed, it needs Torch 2.1.0 to work. Maybe specify the exact version in your requirements. https://softologyblog.wordpress.com/2023/10/10/a-plea-to-all-python-developers/

RealESRGAN and RestoreFormer work fine now too.

Thank you for your feedback. It seems that there have been changes in the api due to torchvision upgrades. I think I can refer to InvokeAI's method invoke-ai/InvokeAI#5690 and add all the dependent code to IOPaint.

carlxwz commented

edit the file: /usr/local/lib/python3.10/dist-packages/basicsr/data/degradations.py
in line 8 change: from torchvision.transforms.functional_tensor import rgb_to_grayscale
to: from torchvision.transforms._functional_tensor import rgb_to_grayscale
Because in Torch 2.2.0 functional_tensor rename to _functional_tensor

  • 👍6
jmaple12 commented

edit the file: /usr/local/lib/python3.10/dist-packages/basicsr/data/degradations.py in line 8 change: from torchvision.transforms.functional_tensor import rgb_to_grayscale to: from torchvision.transforms._functional_tensor import rgb_to_grayscale Because in Torch 2.2.0 functional_tensor rename to _functional_tensor

thanks，I find this file on "D:\Anaconda\Lib\site-packages\basicsr\data\degradations.py",

  • 👍1
lihuaming07 commented

edit the file: /usr/local/lib/python3.10/dist-packages/basicsr/data/degradations.py in line 8 change: from torchvision.transforms.functional_tensor import rgb_to_grayscale to: from torchvision.transforms._functional_tensor import rgb_to_grayscale Because in Torch 2.2.0 functional_tensor rename to _functional_tensor

Thanks, "venv/Lib/site-packages/basicsr/data/degradations.py" may be need for some friends using venv environment.Also, If you are using Pycharm, double click 【shift】 and type "degradations.py", which help you quickly open this file.It's a good means that a would like to mention.

chaoqunxie commented

更新软件包列表并安装所需的软件

RUN apt-get update && apt-get install -y --no-install-recommends
software-properties-common
libsm6 libxext6 ffmpeg libfontconfig1 libxrender1 libgl1-mesa-glx
curl gcc build-essential

升级 pip 并安装特定版本的 torch 和 torchvision

RUN pip install --upgrade pip &&
pip3 install torch==2.1.2 torchvision==0.16.2 --index-url https://download.pytorch.org/whl/cu118

安装 iopaint

RUN pip3 install iopaint

RUN pip install rembg

RUN pip install realesrgan

启动 iopaint 并指定模型、设备和端口以及其他选项

CMD ["iopaint", "start", "--model=lama", "--host=0.0.0.0", "--port=38092", "--enable-interactive-seg", "--enable-remove-bg", "--enable-anime-seg", "--enable-realesrgan", "--enable-gfpgan", "--enable-restoreformer"]

this is my Dockerfile its success