Image & Video Classification for NSFW and SFW media.

Stack:

• FastAPI
• Python
• Docker

Installation

• For ease of use it's recommended to use the provided docker-compose.yml.

version: '3.9'

services:
  image_video_classification:
    image: ghcr.io/doppeltilde/image_video_classification:latest
    ports:
      - "8000:8000"
    volumes:
      - models:/root/.cache/huggingface/hub:rw
    environment:
      - DEFAULT_MODEL_NAME
      - ACCESS_TOKEN
      - DEFAULT_SCORE
      - USE_API_KEYS
      - API_KEYS
    restart: unless-stopped

volumes:
  models:

• Create a .env file and set the preferred values.

DEFAULT_MODEL_NAME=Falconsai/nsfw_image_detection
DEFAULT_SCORE=0.7
ACCESS_TOKEN=

# False == Public Access
# True == Access Only with API Key
USE_API_KEYS=False

# Comma seperated api keys
API_KEYS=abc,123,xyz

Models

Any model designed for image classification and compatible with huggingface transformers should work.

Examples

• https://huggingface.co/Falconsai/nsfw_image_detection
• https://huggingface.co/LukeJacob2023/nsfw-image-detector
• https://huggingface.co/nateraw/vit-age-classifier

Usage

Note

Please be aware that the initial classification process may require some time, as the model is being downloaded.

Tip

Interactive API documentation can be found at: http://localhost:8000/docs

Simple Classification

Image Classification

POST request to the /api/image-classification endpoint.

curl -X 'POST' \
  'http://localhost:8000/api/image-classification?model_name=Falconsai/nsfw_image_detection' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'file=@file.jpg'

Multi Image Classification

POST request to the /api/multi-image-classification endpoint.

curl -X 'POST' \
  'http://localhost:8000/api/multi-image-classification?model_name=Falconsai/nsfw_image_detection' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'files=@0.PNG;type=image/png' \
  -F 'files=@1.JPG;type=image/jpeg' \
  -F 'files=@1.gif;type=image/gif'

Query Classification

You can utilize query parameters, if the standard classification isn't sufficient or you need a more nuanced response.

Optional parameters:

• model_names List[str]

• labels List[str]

• score float

• return_on_first_matching_label bool (default: false)

• fast_mode bool (default: false)

• skip_frames_percentage int (default: 5)

Single Image with Query Parameters

POST request to the /api/image-query-classification endpoint.

curl -X 'POST' \
  'http://localhost:8000/api/image-query-classification?model_names=Falconsai%2Fnsfw_image_detection' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'file=@1.jpeg;type=image/jpeg'

Multi Image with Query Parameters

POST request to the /api/multi-image-query-classification endpoint.

curl -X 'POST' \
  'http://localhost:8000/api/multi-image-query-classification?model_names=Falconsai%2Fnsfw_image_detection' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'files=@1.png;type=image/png' \
  -F 'files=@2.jpeg;type=image/jpeg' \
  -F 'files=@3.gif;type=image/gif'

Video Classification

Video with Query Parameters

POST request to the /api/video-classification endpoint.

curl -X 'POST' \
  'http://localhost:8000/api/video-classification?model_names=Falconsai%2Fnsfw_image_detection' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'file=@1.mp4;type=video/mp4'

Tip

You can find code examples in the examples folder.

---

Notice: This project was initally created to be used in-house, as such the development is first and foremost aligned with the internal requirements.