#tensorflow #tract #onnx

nsfw

A Rust library for detecting NSFW images using a neural network

2 releases

0.1.1 Nov 6, 2023
0.1.0 Nov 5, 2023

#642 in Machine learning

Download history 5/week @ 2023-11-08 7/week @ 2023-11-15 133/week @ 2023-11-22 10/week @ 2023-11-29 28/week @ 2023-12-06 10/week @ 2023-12-13 16/week @ 2023-12-20 4/week @ 2023-12-27 10/week @ 2024-01-03 7/week @ 2024-01-10 6/week @ 2024-01-17 11/week @ 2024-01-24 2/week @ 2024-01-31 6/week @ 2024-02-07 70/week @ 2024-02-14 63/week @ 2024-02-21

143 downloads per month

MIT/Apache

130KB
227 lines

NSFW?

crates.io Documentation Build Status Discord

nsfw is a library for determining the not-safe-for-work-ness of images. It is based off of GantMan's model.

Prerequisites

Because Cargo has a size limit of 10mb, the pre-trained model cannot be included in the crate. You will need to download it from the release page or download the model yourself and convert it into ONNX

Downloading from the release page (easy)

$ gh release download -R Fyko/nsfw --pattern "model.onnx"
# or naviate to the release page and download it manually

Convering from GantMan's model

See our GitHub Workflow

Example: Static Images

[dependencies]
image = { version = "0.24.7", default-features = false, features = ["jpeg"] }
nsfw = { version = "0.1.0", default-features = false, features = ["jpeg"] }
use nsfw::{create_model, examine};

fn main() -> Result<(), Box<dyn std::error::Error>> {
	let model = create_model(
		include_bytes!(concat!(env!("CARGO_MANIFEST_DIR"), "/model.onnx"))
	);

	let image = image::open(concat!(env!("CARGO_MANIFEST_DIR"), "/might_be_porn.jpg"))?;
	let result = examine(&model, &image)?;
	println!("{:#?}", result);

	Ok(())
}

Example: GIF

[dependencies]
image = { version = "0.24.7", default-features = false, features = ["gif"] }
nsfw = { version = "0.1.0", default-features = false, features = ["gif"] }
use nsfw::{examine, MODEL, GifParser};

fn main() -> Result<(), Box<dyn std::error::Error>> {
	let model = create_model(
		include_bytes!(concat!(env!("CARGO_MANIFEST_DIR"), "/model.onnx"))
	);

	let path = concat!(env!("CARGO_MANIFEST_DIR"), "/might_be_porn.gif");
	let file = BufReader::new(File::open(gif_path)?);
	let frames = GifParser::new(GifDecoder::new(file)?, &model);

	for frame in frames {
		println!("{frame:#?}")
	}

	Ok(())
}

Feature Flags

  • default - jpeg and png
  • serde - Enables serialization and deserialization of the model using serde.
  • gif - Enables GIF support for image.
  • jpeg - Enables JPEG support for image.
  • png - Enables PNG support for image.
  • webp - Enables WEBP support for image.

Benchmarking

Name Size Language Time
test_porn.gif 50495726 bytes (50.49 MB) Rust 22.719 seconds
test_porn.gif 50495726 bytes (50.49 MB) JavaScript 219.96 seconds

Dependencies

~25–39MB
~478K SLoC