Hierarchical Layout-Aware Graph Convolutional Network for Unified Aesthetics Assessment

[Paper] [Supplementary Material] [Jittor Code] [Pytorch Code(coming soon)]

This repository contains a Jittor implementation of the paper "Hierarchical Layout-Aware Graph Convolutional Network for Unified Aesthetics Assessment" (CVPR 2021)

Contents

• Requirements
• Install
• Dataset
• Training
• Testing
• Citation

Requirements

Jittor environment requirements:

• System: Linux(e.g. Ubuntu/CentOS/Arch), macOS, or Windows Subsystem of Linux (WSL)
• Python version >= 3.7
• CPU compiler (require at least one of the following)
  • g++ (>=5.4.0)
  • clang (>=8.0)
• GPU compiler (optional)
  • nvcc (>=10.0 for g++ or >=10.2 for clang)
• GPU library: cudnn-dev (recommend tar file installation, reference link)

Install

• Clone repo

  git clone https://github.com/shedy-pub/hlagcn-jittor
  cd hlagcn-jittor
  

• Install dependencies ( jittor, imageio, scikit-learn, opencv-python, pandas. Recommend to use Anaconda.)

  # Create a new conda environment
  conda create -n menv python=3.8
  conda activate menv
  
  # Install other packages
  pip install -r requirements.txt
  

Dataset

• AVA dataset

  • Download the original AVA dataset and dataset spilt into path_to_AVA/. The directory structure should be like:

  path_to_AVAdataset
  ├──aesthetics_image_list
  ├──images
  ├──AVA.txt
  ├──trian.txt
  └──val.txt
  

• AADB dataset

  • Download the AADB dataset into path_to_AADB/. The directory structure should be like:

  path_to_AADBdataset
  ├──AADB_imgListFiles_label
  ├──datasetImages_originalSize
  └──AADB_AllinAll.csv
  

Training

Traning scripts for two datasets can be found in scripts/. The dataroot argument should be modified to path_to_<dataset_name>. Run the follwing command for training:

# Training on AVA
sh script/train_jittor_aadb.sh

# Training on AABD
sh script/train_jittor_aadb.sh

Our code will process the dataset information and save file in preprocess/, which needs few minutes for the first time.

Testing

Testing model by runing the scripts or the follwing command:

python -m utils_jittor.eval \
    --dataset <dataset_name> \
    --dataroot path_to_<dataset_name> \
    --eval_model path_to_model

Citation

If you found this code useful please cite our work as:

@InProceedings{She_2021_CVPR,
    author    = {She, Dongyu and Lai, Yu-Kun and Yi, Gaoxiong and Xu, Kun},
    title     = {Hierarchical Layout-Aware Graph Convolutional Network for Unified Aesthetics Assessment},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2021},
    pages     = {8475-8484}
}