This project provides a pretrained image classifier for detecting lewd content, based on Bumble's internal model. It's designed for developers and researchers needing to implement or fine-tune NSFW image detection, offering a robust solution built on EfficientNet-v2.

How It Works

The model utilizes an EfficientNet-v2 architecture, trained on a proprietary dataset of lewd images. This approach leverages transfer learning for efficient and accurate classification of potentially sensitive content. The model is released in SavedModel format, facilitating straightforward deployment across various platforms.

Quick Start & Requirements

• Install: Use conda env create -f environment.yaml and conda activate private_detector.
• Prerequisites: Python, Conda.
• Inference: Run python3 inference.py --model saved_model/ --image_paths <path_to_images>.
• Resources: Requires downloading the private_detector.zip archive containing SavedModel and checkpoint files.
• Docs: Whitepaper

Highlighted Details

• Pretrained SavedModel and Frozen Model available.
• Supports fine-tuning on custom datasets.
• Includes example inference script and TensorFlow Serving example.
• Offers various training parameters for customization.

Maintenance & Community

This is an open-source release from Bumble. Further community engagement details (e.g., Discord, Slack) are not specified in the README.

Licensing & Compatibility

The README does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The model is trained on an internal dataset, and its performance on diverse datasets may vary. The specific dataset composition and potential biases are not detailed.