dreadlord1984 / nima.pytorch

NIMA: Neural IMage Assessment

PyTorch NIMA: Neural IMage Assessment

PyTorch implementation of Neural IMage Assessment by Hossein Talebi and Peyman Milanfar. You can learn more from this post at Google Research Blog.

Installing

Docker

docker run -it truskovskiyk/nima:latest /bin/bash

PYPI package (In Progress)

pip install nima

VirtualEnv

git clone https://github.com/truskovskiyk/nima.pytorch.git
cd nima.pytorch
virtualenv -p python3.7 env
source ./env/bin/activate

Dataset

The model was trained on the AVA (Aesthetic Visual Analysis) dataset You can get it from here Here are some examples of images with theire scores

Pre-train model (In Progress)

Deployment (In progress)

Usage

nima-cli

Usage: cli.py [OPTIONS] COMMAND [ARGS]...

Options:
  --help  Show this message and exit.

Commands:
  get_image_score  Get image scores
  prepare_dataset  Parse, clean and split dataset
  run_web_api      Start server for model serving
  train_model      Train model
  validate_model   Validate model

Previous version of this project is still valid and works

you can find here

Contributing

Contributing are welcome

License

This project is licensed under the MIT License - see the LICENSE file for details

Acknowledgments

• neural-image-assessment in keras
• Neural-IMage-Assessment in pytorch
• pytorch-mobilenet-v2
• origin NIMA article
• origin MobileNetV2 article
• Post at Google Research Blog
• Heroku: Cloud Application Platform