[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)
Jittor environment requirements:
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
AVA dataset
path_to_AVA/
. The directory structure should be like:path_to_AVAdataset
├──aesthetics_image_list
├──images
├──AVA.txt
├──trian.txt
└──val.txt
AADB dataset
path_to_AADB/
. The directory structure should be like:path_to_AADBdataset
├──AADB_imgListFiles_label
├──datasetImages_originalSize
└──AADB_AllinAll.csv
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 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
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}
}