Introduction

Finally some good profile pictures! Hugging spaces, for testing purposes only.

Got tired of constantly searching for new profile pictures? Or maybe even just the thought of changing it is a hassle.

Well, Fsg_Pp aims to automate that for you! Just type what you want to find and it will filter out the best ones for you!

Long awaited gallery isolation and asynchronous search between tabs are released!

Quick Links

• Installing and Running
• Features
  • AI Mode
  • Automatic Crop
  • Image Search
  • Pixiv
  • Danbooru
  • Zerochan
• Manual Installation
  • Windows
  • macOS/Linux
• Questionable Ideas
• Troubleshooting
• Acknowledgment
• Disclaimer

Installing and Running

1. Install Python (Recommend 3.10.7 and higher), while checking "Add Python to PATH"

2. Install git

   Note Skip steps 2 and 3 if you would like to download the release here Fsg-Pp releases

3. Clone the repo git clone https://github.com/EngMarchG/Fsg_Pp.git

   Note If you prefer to install the requirements manually, please head to Manual Installation

4. Run the installation.bat(Windows)/installation.sh(macOS/linux) and it will set everything up for you!

   Warning to run installation.sh and launcher.sh on macOS/linux, run the following in your terminal at the current folder containing the scripts; else scripts will fail:

   To give permissions for the scripts to run:

    chmod 755 installation.sh
    chmod 755 launcher.sh
   

   Finally, to run the scripts:

    ./installation.sh
    ./launcher.sh
   

5. After installation.bat(Windows)/installation.sh(macOS/linux) has successfully finished, run launcher.bat(Windows)/launcher.sh(macOS/linux)

   Note To close the script process, press CTRL+C

Features

AI Mode

Uses a pretrained model to classify images suitable to be used as a profile picture. Note: AI mode usually attempts to classify images based on the thumbnail (this is beneficial for both the website and the user)

Automatic Crop

Automatically detect faces and crop your images! Just drag and drop or click to upload an image.

Image Search

This app currently allows you to search for images using three sites:

1. Pixiv
2. Danbooru
3. Zerochan

Pixiv

• Allows the searching of preview premium images and free images
• Restricts the search type to SFW and NSFW images
  • If you are not logged in pixiv only offers SFW images
  • Default queries are based on account settings, so be sure R-18 is enabled if you want to use it
• Filter by Likes, bookmarks and/or views
• Download images in the standard or native size
• Continue on the previous page it ended or if it crashes (Ignores current query when checked)

Note: Pixiv search will ignore your option filters if you are not logged in. You only have to log in once, unless you delete the browser profile created in the commands folder.

Danbooru

• Allows downloading images that are ordered by Score
• Allows filtering by tags (both inclusions and exclusions)
• Has 3 modes of image restrictions gradually increasing the PG-Friendliness of the images found (More Pg < Sensitive < Strictly PG)
• Continue on the previous page it ended or if it crashes (Ignores current query when checked)

Zerochan

• Filter by Likes
• Continue on the previous page it ended or if it crashes (Ignores current query when checked)

Note: It is better to make the query specific such as using the character's full name. Ex: Searching 'Honkai' would fail due to the existance of several similar names. Searching 'Honkai star rail', on the other hand, would work since it is more specific and wouldn't get mixed up with other names.

Yandex

• Filter by Image orientation and recent upload date (up to past 7 days)

Manual Installation

To install the requirements manually on Windows, please use the following commands:

Windows

Right click in the installed folder and open a new windows terminal

Activate the Python environment:

python -m venv venv

Activate the virtual environment:

venv\Scripts\activate.bat

Install dependencies from requirements.txt:

venv\Scripts\python.exe -m pip install -r requirements.txt

For GPU torch (NVIDIA GPUs Only, ROCm is currently not supported due to its limited support from pytorch):

venv\Scripts\pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

venv\Scripts\pip install ultralytics

For CPU torch (Default):

venv\Scripts\pip install ultralytics

Check if cv_files folder exists, create if it doesn't:

if not exist cv_files (
mkdir cv_files
)

Install AniClassifier.pt and AniFaceDet.pt inside cv_files folder:

cd cv_files

Install AniClassifier.pt and AniFaceDet.pt inside cv_files folder:

curl -L -o AniClassifier.pt https://huggingface.co/datasets/Kyo-Kai/Fsg_pp_files/resolve/main/AniClassifier.pt 

curl -L -o AniFaceDet.pt https://huggingface.co/datasets/Kyo-Kai/Fsg_pp_files/resolve/main/AniFaceDet.pt

To install the requirements manually on macOS/Linux, please use the following commands:

macOS/Linux

Create the Python virtual environment:

python3 -m venv venv

Activate the Python environment:

source venv/bin/activate

Install dependencies from requirements.txt:

python3 -m pip install -r requirements.txt

Install PyTorch with GPU support on Linux:

(NVIDIA GPUs):

python3 -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

python3 -m pip install ultralytics

(AMD GPUs):

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.4.2

python3 -m pip install ultralytics

Install Ultralytics using pip (macOS):

python3 -m pip install ultralytics

Install AniClassifier.pt and AniFaceDet.pt inside cv_files folder:

cd cv_files
curl -L -o AniClassifier.pt https://huggingface.co/datasets/Kyo-Kai/Fsg_pp_files/resolve/main/AniClassifier.pt
curl -L -o AniFaceDet.pt https://huggingface.co/datasets/Kyo-Kai/Fsg_pp_files/resolve/main/AniFaceDet.pt

Go back to the parent directory

cd ..

Deactivate the virtual environment

deactivate

Note GPU torch on macOS requires macOS 12.3 or later, for more info: Accelerated PyTorch training on Mac

Questionable Ideas

• Using NLP for better tag searching/filtering (particularly for danbooru)

• Pixiv: Search for a specific Artist's artworks

Troubleshooting

For Windows users, in case you have problems with the installation.bat, disable the aliases in Windows:
If you are able to use python in the cmd, but can't run the script, then you probably have permission issues. In that case, refer to the manual installation guide or you can try to change your execution policies (not recommended)

For macOS/Linux users, in case you face problems while running both launcher.sh and installation.sh, make sure the proper access is granted to these files by running the following commands in a terminal at the same folder of both files:

chmod 755 launcher.sh
chmod 755 installation.sh

Then run the following to launch the scripts:

./installation.sh
./launcher.sh

Note In case selenium is unable to locate a driver, put your custom chrome driver in the driver folder. The chrome driver must not exceed your installed chrome version.

Acknowledgment

We would like to thank the authors of the following datasets for contributing to a portion of the face detector model's dataset:

Authors of the Anime Heads Dataset

Authors of the Face Detection Dataset

Special thanks to gradio for providing an intuitive and flexible UI.

Disclaimer

The main objective of the application is, as the name implies, to download pictures that are suitable for use as profile pictures.

Therefore, we will NOT be incorporating any type of multithreaded functionality.

That said, users can still expect some performance improvements and optimizations.

Furthermore, if you appreciate the artists' artwork, please consider following their respective accounts. The chosen naming convention makes it easy to trace back to the artists.