NudeNet NSFW Tagging API

A simple REST API built with Flask and NudeNet for detecting and tagging NSFW content in images.
Supports file uploads and remote image URLs (single or multiple).

---

Features

• 🔥 Fast inference — NudeNet model preloaded at container startup
• 🖼️ Flexible input — Upload files or pass one/many image URLs
• ⚡ Production-ready — Gunicorn + Dockerfile included
• 🛡️ Safe defaults — File type validation, download timeouts
• ✅ Health check — /health endpoint for container orchestration

---

API Endpoints

Health check

GET /health

Returns:

{ "status": "ok" }

Detect NSFW (single file)

POST /detect
Content-Type: multipart/form-data
Form field: file=@/path/to/image.jpg

Example:

curl -X POST http://localhost:8080/detect \
  -F "file=@/path/to/image.jpg"

Detect NSFW (single URL)

POST /detect
Content-Type: application/json
Body:
{
  "image_url": "https://example.com/image.jpg"
}

---

Responses

{
  "ok": true,
  "mode": "urls",
  "count": 2,
  "results": [
    {
      "source": "https://example.com/a.jpg",
      "detections": [
        { "class": "EXPOSED_ANUS", "score": 0.98, "box": [0.2, 0.3, 0.5, 0.7] }
      ]
    },
    {
      "source": "https://example.com/b.png",
      "error": "HTTP 404"
    }
  ]
}

• detections: each detection contains class (NSFW label), score (confidence), and box (normalized bounding box [x_min, y_min, x_max, y_max]).
• error: returned if the file/URL failed.

---

Local Development

1️⃣ Clone & install

git clone https://github.com/yourusername/nsfw-tagging-api.git
cd nsfw-tagging-api
pip install -r requirements.txt

2️⃣ Run (dev mode)

python app.py
# or specify port:
python app.py --port 8080

Server runs at http://127.0.0.1:8080.

---

Docker

Build

docker build -t nsfw-tagging-api .

Run

docker run --rm -p 8080:8080 nsfw-tagging-api

Test

curl -X POST http://127.0.0.1:8080/detect \
  -F "file=@/path/to/image.jpg"

---

Environment

• Python 3.11+
• Flask + Flask-RESTful
• NudeNet (downloads pretrained models)
• Requests

Ensure your container/machine has enough memory (>=512 MB recommended).

---

License

MIT © 2025 [Hrishikesh Kanade]