Skip to content

Latest commit

 

History

History
400 lines (294 loc) · 29.6 KB

README_en.md

File metadata and controls

400 lines (294 loc) · 29.6 KB

Here's the English translation of the provided text:

NoneBotPluginLogo

NoneBotPluginText

Nonebot-Plugin-NaiLongRemove

✨ A simple plugin based on an AI model~ ✨

python pdm-managed
Pydantic Version 1 Or 2 license pypi pypi download
NoneBot Registry Supported Adapters

简体中文 | English |

📖 Introduction

Disclaimer

This plugin is for entertainment and educational purposes only.

Overview

NaiLongRemove is a simple AI model-based plugin designed to identify "NaiLong" meme images in groups and automatically remove them.

Technology

Currently, the plugin supports three models, which can be switched via the configuration file, as detailed in the configuration section below.
Users can choose the model that best suits their needs; both models have been optimized for performance, but errors may still occur to varying degrees. Feedback is welcome!

💿 Installation

Cloud Deployment

  • The run_with_napcat_en.ipynb file supports one-click deployment on platforms like Kaggle or Huggingface's Spaces. You can complete the bot deployment simply by clicking "Run" and scanning the QR code!
  • Supports Docker one-click deployment.

Local Deployment

If you are new to NoneBot, please refer to this guide

To avoid dependency issues, we've separated the installation methods for using GPU inference from the regular installation. Choose the one that suits your needs.

Using CPU Inference

You can choose one of the following methods.

[Recommended] Install using nb-cli Open a terminal in the root directory of the NoneBot2 project and run the following command to install: 
nb plugin install nonebot-plugin-nailongremove
Install using a package manager Open a terminal in the plugin directory of the NoneBot2 project, and use the appropriate command based on your package manager.
pip 
pip install nonebot-plugin-nailongremove
pdm 
pdm add nonebot-plugin-nailongremove
poetry 
poetry add nonebot-plugin-nailongremove
conda 
conda install nonebot-plugin-nailongremove

Open the pyproject.toml file in the root directory of the NoneBot2 project, and add the following to the [tool.nonebot] section:

[tool.nonebot]
plugins = [
    # ...
    "nonebot_plugin_nailongremove"
]

Using GPU Inference

Click to expand

[!NOTE] These steps are more technical and complicated, so non-technical users may choose to skip them.
In fact, CUDA acceleration has minimal impact on the models used by this plugin, so do not attempt these steps without understanding them.

First, enter the Bot virtual environment (if any).

[!NOTE] If you've previously installed the CPU inference version, please uninstall it first:

pip uninstall nonebot-plugin-nailongremove torch torchvision onnxruntime

Install the base package:

pip install nonebot-plugin-nailongremove-base

Based on your installed CUDA and CuDNN versions (if you have them; if not, install them), follow the official instructions to install the corresponding dependencies:

After installation, configure the plugin to use CUDA for inference:

NAILONG_ONNX_PROVIDERS=["CUDAExecutionProvider"]

Finally, configure NoneBot2 to load the plugin. Open the pyproject.toml file in the root directory of the NoneBot2 project, and add the following to the [tool.nonebot] section:

[tool.nonebot]
plugins = [
    # ...
    "nonebot_plugin_nailongremove"
]

When updating the plugin later, you only need to update the base package. Do not install or update the non-base package:

pip install nonebot-plugin-nailongremove-base -U

⚙️ Configuration

Add the required configurations from the table below to the .env file in your NoneBot2 project.

Configuration Item Required Default Value Description
Global Configuration
PROXY No None Proxy address used for downloading model files.
Response Configuration
NAILONG_BYPASS_SUPERUSER No False Whether to bypass checking images sent by superusers.
NAILONG_BYPASS_ADMIN No False Whether to bypass checking images sent by group administrators.
NAILONG_NEED_ADMIN No False Whether to skip checking all images in the group when the bot is not an administrator.
NAILONG_LIST_SCENES No [] List of allowed or disallowed chat scene IDs (e.g., group IDs or channel IDs).
NAILONG_BLACKLIST No True Whether to use blacklist mode.
NAILONG_USER_BLACKLIST No [] List of user IDs to be blacklisted.
NAILONG_PRIORITY No 100 Matcher priority.
Behavior Configuration
NAILONG_RECALL No ["nailong"] Whether to recall the message
NAILONG_MUTE_SECONDS No {"nailong":0} Set the mute duration. If not set or the duration is 0, no mute will be applied.
Unit: seconds
NAILONG_TIP No {"nailong": ["This group prohibits NaiLong images!"]} Message to send as a tip, using Alconna message template with custom variables.
NAILONG_FAILED_TIP No {"nailong": ["{:Reply($message_id)} Oh no, please don't send NaiLong images! 🥺 👉👈"]} Message sent when recalling fails or when recalling is disabled.
NAILONG_CHECK_ALL_FRAMES No False Specifies whether to check all frames in the image when using model 1. Requires setting NAILONG_CHECK_MODE to 0. When enabled, the $checked_result variable in the message template will return a GIF if the original image is animated.
NAILONG_CHECK_RATE No 0.8 When checking all frames of an image, the image will only be recalled or processed if a certain proportion of the frames meet the detection criteria.
NAILONG_CHECK_MODE No 0 Selects the detection method for GIF animations.
0. Check all frames
1. Check only the first frame
2. Random frame sampling
Similarity Detection Configuration
NAILONG_SIMILARITY_ON No False Specifies whether to enable similarity detection on local storage before processing images.
NAILONG_SIMILARITY_MAX_STORAGE No 10 Maximum number of errored images stored locally. When the limit is reached, older images will be compressed and uploaded to the database, but it won’t affect previously stored images.
NAILONG_HF_TOKEN No None Hugging Face Access Token, which allows automatic data upload to Hugging Face and becomes a contributor to the dataset.
General Model Configuration
NAILONG_MODEL_DIR No ./data/nailongremove The download location for the model.
NAILONG_MODEL No 1 Specifies which model to load; available models are listed below.
NAILONG_AUTO_UPDATE_MODEL No True Specifies whether to automatically update the model.
NAILONG_CONCURRENCY No 1 Maximum concurrency number for identifying frames of animated images.
NAILONG_ONNX_PROVIDERS No ["CPUExecutionProvider"] List of providers used for loading ONNX models; please refer to installation documentation above.
Model 1 Specific Configuration
NAILONG_MODEL1_TYPE No tiny The model type used for model 1; options are tiny / m.
NAILONG_MODEL1_YOLOX_SIZE No None For model 1, custom input sizes may be changed.
Model 2 Specific Configuration
NAILONG_MODEL2_ONLINE No False For model 2, specifies whether to enable online inference; this mode is currently not suitable with NAILONG_CHECK_MODE set to 0.
Model 1&2 Specific Configuration
NAILONG_MODEL1_SCORE No {"nailong": 0.5} Confidence threshold for models 1&2; range is 0 ~ 1. Values can be customized according to labels; set the corresponding threshold for detection, setting it to null or leaving it blank will ignore that label.
Miscellaneous Configuration
NAILONG_GITHUB_TOKEN No None GitHub Access Token; can be filled in to address issues with model downloads or updates.

Available Models

Available Variables in Message Templates

Variable Name Type Description
$event Event The current event
$target Target The target of the event
$message_id str The message ID
$msg UniMessage The current message
$ss Session The current session
$checked_result Image The image after selecting the corresponding target, exists only when the model is configured as 1.

🎉 Usage

Whenever a "NaiLong" meme is recognized, it will be retracted and notified.

To store errored images locally (for SUPERUSERS): Send "This is [type]"+image, for example: "This is nailong+image", and it will be automatically stored locally. When similarity detection is on, in the next image check, it will prioritize recognizing the images already stored locally.

📞 Contact

Welcome everyone to join the group for learning and exchange!

📝 Changelog

2.3.6

  • Fixed some SSL connection errors

Minor Update

  • Added the run_with_napcat_en.ipynb file, which supports one-click deployment on platforms like Kaggle or Huggingface's Spaces. You can complete the bot deployment simply by clicking "Run" and scanning the QR code!
  • Added Docker one-click deployment.

2.3.5

  • The update adds a feature to select a mute tag, allowing users to choose whether to mute or recall the processing for different types of images.
  • A new configuration option, NAILONG_CHECK_RATE, has been added. When detecting all frames of an animated image, this optional configuration allows success in judgment when the proportion of frames containing the "nailong" frame reaches a certain threshold.

2.3.4

  • Added model3 to NAILONG_MODEL, a model trained based on YOLOv11. It is recommended to set {"nailong": 0.78} in the NAILONG_MODEL1_SCORE configuration, and NAILONG_MODEL1_YOLOX_SIZE to [640,640].
  • Updated default values for configuration items: NAILONG_BYPASS_SUPERUSER -> False, NAILONG_BYPASS_ADMIN -> False.

2.3.3

  • Optimized temporary processing solutions to reduce performance pressure and improve speed (the vector library faiss also supports GPU processing, but it is not recommended for non-professionals to use GPU due to the complex installation process).
  • Added NAILONG_HF_TOKEN to automatically upload errored images to the Hugging Face dataset.
  • Changed the formats of the configuration items NAILONG_TIP and NAILONG_FAILED_TIP, allowing random response messages. When the corresponding value is an empty list [], only the image will be checked (or the mute/revoke action will be performed) without returning a message.

2.3.2

  • Updated the three frame processing modes for GIFs. You can choose through NAILONG_CHECK_MODE.
  • Updated the temporary handling for errored images. By enabling NAILONG_SIMILARITY_ON, local storage similarity matching can be used. Additionally, by sending "This is [type]"+image through SUPERUSERS, errored images can be saved to local records.
  • Added model2 to NAILONG_MODEL, which is based on the YOLOv11-trained model. Currently, it only supports Nailong recognition.

2.3.1

  • Modified plugin dependencies to avoid some issues that affected the installation process. Please refer to the installation documentation for more details.
    • Corresponding configuration changes: Removed the NAILONG_ONNX_TRY_TO_USE_GPU configuration item and added the NAILONG_ONNX_PROVIDERS configuration item.

2.3.0

  • Added support for checking all frames in a GIF and re-encapsulating the results into a new GIF. This is disabled by default. The $checked_image variable has been deprecated, and a new $checked_result variable has been added.
  • The input size for model 1 can now be automatically configured based on the model type, but if specified in the configuration, it will be used as the priority.
  • Supported processing of images containing other tags. Some configuration items now allow custom values based on the tags.
  • Added a user blacklist.
  • The default model has been changed to 1.

2.2.1

  • Optimized model auto-update (possibly a reverse optimization).

2.2.0

  • Renamed the configuration item NAILONG_YOLOX_SIZE to NAILONG_MODEL1_YOLOX_SIZE.
  • Model 1 can now automatically get the latest version, and you can also choose the model type through configuration.
  • Model 1 can now control the confidence threshold for recognition via configuration.
  • When loading the ONNX model, the system will attempt to use GPU by default. If it fails, a warning will be shown. If you don't want to see the warning, you can refer to the above to disable the corresponding configuration.

2.1.4

  • Fixed a bug where the NAILONG_NEED_ADMIN configuration was not effective.

2.1.3

  • Fixed a bug where the group manager and superuser settings were ignored.

2.1.2

  • Refactored some code and fixed potential bugs.

2.1.1

  • Added the $checked_image variable.

2.1.0

  • Download models from the original repository.

2.0.0

  • Refactored the plugin to support multiple platforms.
  • Updated two new models and optimized model accuracy. Users can choose one of them for inference.
  • Added features like mute, group blacklist and whitelist, optional admin detection disablement.
  • Added an option for auto-updating the model.