GEneral Audience Content Classifier (Geacc) is a pre-trained deep neural network model for classifying images which are not suitable for general audiences.
Currently, the model detects sexually explicit content. In the future, we plan to extend it to other types of harmful visual content such as violence and horror.
If you find it useful, please let us know about your use case by filling in short form here. As a non-profit organisation, it's essential for us to gauge the impact of our work.
Pre-trained model is available in Keras HDF5 format and can be downloaded here.
Definition of "General Audience" varies depending on the country and type of content. Our definition influenced by Television Content Rating systems which are stricter than movie and gaming rating systems. TV content rating systems of many countries define "General Audience" content suitable for children under 12.
The precise definition of sexually explicit content is also highly subjective. In this project we consider any visual material that may cause sexual arousal or fantasy, whether intentional or unintentional, to be harmful. Framing the problem in that particular way prioritises child safety over objectivity, removes ambiguity and makes machine learning model more robust.
To classify images using Geacc model, you can use Docker image which includes all dependencies such as TensorFlow. Alternatively, if you already have TensorFlow setup, you can use utils/predict.py script provided in this repo.
To classify all images in the current directory using Docker image:
docker run -it --rm --volume=$(pwd):/myimages purifyai/geacc:latest /geacc/predict.py -m /geacc/PurifyAI_Geacc_MobileNetV2_224.h5 /myimages/
Usage:
% utils/predict.py -h
usage: predict.py [-h] -m MODEL_PATH [-c CSV_PATH] input_path
positional arguments:
input_path Path to the input image or folder containing images.
optional arguments:
-h, --help show this help message and exit
-m MODEL_PATH, --model_path MODEL_PATH
Path to the trained Keras model in HDF5 file format
-c CSV_PATH, --csv_path CSV_PATH
Optional. If specified, results will be saved to CSV
file instead of stdout
Example:
% utils/predict.py -m ./models/PurifyAI_Geacc_MobileNetV2_224.h5 /path/to/myimages/
Using TensorFlow backend.
Filename Prediction Latency (ms) Benign Score Malign Score
0 ../img/dog1.jpg benign 62 0.7445 0.2555
1 ../img/nsfw1.jpg malign 89 0.0000 1.0000
2 ../img/nsfw2.jpg malign 91 0.0150 0.9850
3 ../img/cat1.jpg benign 74 0.8820 0.1180
This model based on lightweight MobileNetV2 architecture which was specifically designed to run on personal mobile devices.
This choice aligned with our vision for an on-device child protection systems which we described here.
In the future, we plan to release pre-trained models using other architectures which may provide higher accuracy for those who don't need to run inference on mobile devices (e.g. Inception).
The dataset consists of 50.000 images, equally split between two classes - benign and malign. In addition to that, test dataset of ~3.600 images used for validation.
| Class | Training Images | Test Images |
|---|---|---|
| Benign | 25,000 | 1,800 |
| Malign | 25,000 | 1,800 |
Images in this dataset are mainly collected online, while some images taken from Caltech256 and Porn Database.
We do not provide original dataset due to the nature of the data and the fact that Purify Foundation does not own the copyright of those images. However, in the future, we plan to publish the list of URLs.
Training was performed using Keras with TensorFlow backend.
Instead of training from scratch, we used fine-tuning approach. MobileNetV2 model pre-trained with ImageNet was used as the basis for the Geacc model. The top fully connected layer was replaced for binary classification (malign/benign). All layers except the top 30 layers were frozen.
Images were resized and cropped to match 224x224 input size and augmented to improve accuracy.
Hyperparameters and other details can be found in the source code of the training script.
Geacc MobileNetV2 model achieved more than 85% accuracy on the test dataset. We are working to improve these results.
This project is currently in the early development stage. We do not provide guarantees of output accuracy.
Models and source code are licensed under Apache License 2.0