yzc002/TMCR

Semantics-Aware Image Aesthetics Assessment using Tag Matching and Contrastive Ranking

Zhichao Yang¹, Leida Li^1*, Pengfei Chen¹, Jinjian Wu¹, Weisheng Dong¹,

¹School of Artificial Intelligence, Xidian University

^*Corresponding author

Introduction：

PyTorch implementation for the paper

Model weight：./model

Inference Guide：

1. Overview

This guide will help you get started with the TMCR inference code.

2. Directory Structure

   project_root/
├── AVA/
│   └── Image/                       # AVA dataset images
│   └── Label/                       # AVA dataset labels
├── TM/
│   └── Attr_Tags.csv                 # Aesthetic attribute Tags
│   └── Attr_Tags.csv                 # Sementic attribute Tags
│   └── TM_AVA.py                     # Extract TM Features
├── CR/
│   └── CR_AVA.py                     # Extract CR Features (Training)
├── TMCR/
│   └── TMCR.py                       # Testing TMCR on AVA

3. Download Required Files

Swin-B Pretrained Weights: Place in ./Model/swin_b-68c6b09e.pth
TMCR Model: Place your trained model at ./Model/TMCR_AVA.pt
AVA Images: Download AVA dataset images to ./AVA/images/

4. Prepare Test Data

Your test_TM.csv should have the following format:

image_id,score_1,score_2,...,score_10,TM_feature
123456,10,20,30,...,50,"[0.1,0.2,0.3,...,0.9]"
Columns 1-11: Image ID and 10 aesthetic score distributions
Column 12: TM_feature as a string representation of a vector

5. Running Inference

python TMCR.py

Citation

If you find our work is useful, pleaes cite the paper:

@inproceedings{yang2024semantics,  
  title={Semantics-Aware Image Aesthetics Assessment using Tag Matching and Contrastive Ranking},  
  author={Yang, Zhichao and Li, Leida and Chen, Pengfei and Wu, Jinjian and Dong, Weisheng},  
  booktitle={Proceedings of the 32nd ACM International Conference on Multimedia},  
  pages={2632--2641},  
  year={2024}  
}