m1chlcz / nsfw_sherlock

Simple drop in API to determine if image is NSFW using TensorFlow

NSFW Sherlock

Build with GO and TensorFlow

Simple drop-in API for classifying images for NSFW content.

It can determine if picture is NSFW as well as if text on picture is NSFW

Both HTTP and GRPC APIs are supported

Use docker to run NSFW Sherlock

UPDATE 0.0.0.4:

Now you can use additional API, which will give you labels (Drawings, Neutral, Hentai, Porn, Sexy) and its values directly. Implemented on both Web and GRPC APIs. For Web: /pic/labels For GRPC: Please refer to protofile in ./proto

Usage with Git Clone:

git clone https://github.com/M1chlCZ/nsfw_sherlock nsfw_sherlock
cd nsfw_sherlock
docker build -t nsfwsherlock .

GRPC:
docker run -e APP_ENV=grpc -p 4000:4000 nsfwsherlock 
(proto file in /proto)

HTTP:
docker run -e APP_ENV=http -p 4000:4000 nsfwsherlock
(POST /pic/check) req: {"base64": "base64 string of image", filename: "image.jpg"}
res: {
    "status":   "ok", 
    "message":  "success", 
    "nsfwText": true/false, 
    "nsfwPic":  true/false
}

(POST /pic/labels) req: {"base64": "base64 string of image", filename: "image.jpg"}, 
res: {
    "status":   "ok",
	"message":  "success",
	"drawings": 0.0,
	"hentai":   0.0,
	"neutral":  0.0,
	"porn":     0.0,
	"sexy":     0.0,
	"nsfwText": true/false,
  },

---------------------------------------------

LOADING YOUR OWN BAD WORDS:
docker run -e APP_ENV=grpc/http  -v /path/to/your/host/bad_words.txt:/bad_words.txt -p 4000:4000 nsfwsherlock

If you need tighten up or loosen up NSFW detection rules, you can do so in nsfw/nsfw.go file or use /pic/labels endpoint

Usage with Docker Hub:

docker pull m1chl/nsfw-sherlock

GRPC:
docker run -e APP_ENV=grpc -p 4000:4000 m1chl/nsfw-sherlock 
(proto file in /proto)

HTTP:
docker run -e APP_ENV=http -p 4000:4000 m1chl/nsfw-sherlock
(POST /pic/check) req: {"base64": "base64 string of image", filename: "image.jpg"}
res: {
    "status":   "ok", 
    "message":  "success", 
    "nsfwText": true/false, 
    "nsfwPic":  true/false
}

(POST /pic/labels) req: {"base64": "base64 string of image", filename: "image.jpg"}, 
res: {
    "status":   "ok",
	"message":  "success",
	"drawings": 0.0,
	"hentai":   0.0,
	"neutral":  0.0,
	"porn":     0.0,
	"sexy":     0.0,
	"nsfwText": true/false,
  },


---------------------------------------------

LOADING YOUR OWN BAD WORDS:
docker run -e APP_ENV=grpc/http  -v /path/to/your/host/bad_words.txt:/bad_words.txt -p 4000:4000 nsfwsherlock

Credits:

Model: https://github.com/GantMan/nsfw_model

Tensorflow-GO: https://github.com/galeone/tfgo

OCR: https://github.com/otiai10/gosseract