Rethinking Personalized Aesthetics Assessment: Employing Physique Aesthetics Assessment as An Exemplification

Haobin Zhong, Shuai He, Anlong Ming, Huadong Ma

Beijing University of Posts and Telecommunications

---

[国内的小伙伴可以看这] This repo contains the official implementation and the new dataset PhysiqueAA50K of the CVPR 2025 paper.

Introduction

The Personalized Aesthetics Assessment (PAA) aims to accurately predict an individual's unique perception of aesthetics. With the surging demand for customization, PAA enables applications to generate personalized outcomes by aligning with individual aesthetic preferences. The prevailing PAA paradigm involves two stages: pre-training and fine-tuning, but it faces three inherent challenges:

1. The model is pre-trained using datasets of the Generic Aesthetics Assessment (GAA), but the collective preferences of GAA lead to conflicts in individualized aesthetic predictions.
2. The scope and stage of personalized surveys are related to both the user and the assessed object; however, the prevailing personalized surveys fail to adequately address assessed objects' characteristics.
3. During application usage, the cumulative multimodal feedback from an individual holds great value that should be considered for improving the PAA model but unfortunately attracts insufficient attention.

To address the aforementioned challenges, we introduce a new PAA paradigm called PAA+, which is structured into three distinct stages: pre-training, fine-tuning, and continual learning. Furthermore, to better reflect individual differences, we employ a familiar and intuitive application, physique aesthetics assessment (PhysiqueAA), to validate the PAA+ paradigm. We propose a dataset called PhysiqueAA50K, consisting of over 50,000 annotated physique images. Furthermore, we develop a PhysiqueAA framework (PhysiqueFrame) and conduct a large-scale benchmark, achieving state-of-the-art (SOTA) performance. Our research is expected to provide an innovative roadmap and application for the PAA community.

Code Usage Instructions

• Training Objective

The goal is to predict the PhysiqueAA Scores: appearance, health, and posture.

• Installation

conda create -n physiqueAA python=3.10.14
conda activate physiqueAA
cd ./code
pip install -r requirements.txt
bash script.sh

download SMPLer_X checkpoints from Baidu Netdisk to ./code/SMPLer_X/pretrained_models

download smplx.npz from Baidu Netdisk to .code/SMPLer_X/common/utils_smpler_x/human_model_files/smplx

download Swinv2 checkpoints from Baidu Netdisk to ./code/models_/pam/pretrained

All resources can also download from OneDrive

If you encounter the following error message:

RuntimeError: Subtraction, the `-` operator, with a bool tensor is not supported. If you are trying to invert a mask, use the `~` or `logical_not()` operator instead.

The Solution: Modify 1- mask to ~mask in torchgeometry/core/conversions.py

Example: Modify mask_c2 = (1 - mask_d2) * mask_d0_nd1 to mask_c2 = (~mask_d2) * mask_d0_nd1

• Training Steps

1. Download the dataset from Baidu Netdisk.
2. Use train.py to train the network.

• Inference

1. Use inference.py to inference the network.

• PhysiqueFrame Checkpoint Files

1. Download the PhysiqueFrame checkpoint files from Baidu Netdisk.

Unfortunately, PENet cannot be open-sourced due to its enterprise use. However, we have included the preference feature (preference tensor) generated by PENet in the dataset.

If you find our work is useful, please cite our paper:

@inproceedings{zhong2025rethinking,
  title={Rethinking Personalized Aesthetics Assessment: Employing Physique Aesthetics Assessment as An Exemplification},
  author={Zhong, Haobin and He, Shuai and Ming, Anlong and Ma, Huadong},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={2935--2944},
  year={2025}
}

🎁 Projects	📚 Publication	🌈 Content	⭐ Stars
Attacker Against IAA Model【美学模型的攻击和安全评估框架】	TIP 2025	Code, Dataset
Personalized Aesthetics Assessment【个性化美学评估新范式】	CVPR 2025	Code, Dataset
Pixel-level image exposure assessment【首个像素级曝光评估】	NIPS 2024	Code, Dataset
Long-tail solution for image aesthetics assessment【美学评估数据不平衡解决方案】	ICML 2024	Code
CLIP-based image aesthetics assessment【基于CLIP多因素色彩美学评估】	Information Fusion 2024	Code, Dataset
Compare-based image aesthetics assessment【基于对比学习的多因素美学评估】	ACMMM 2024	Code
Image color aesthetics assessment【首个色彩美学评估】	ICCV 2023	Code, Dataset
Image aesthetics assessment【通用美学评估】	ACMMM 2023	Code
Theme-oriented image aesthetics assessment【首个多主题美学评估】	IJCAI 2022	Code, Dataset
Select prompt based on image aesthetics assessment【基于美学评估的提示词筛选】	IJCAI 2024	Code
Motion rhythm synchronization with beats【动作与韵律对齐】	IJCAI 2024	Code, Dataset
Champion Solution for AIGC Image Quality Assessment【NTIRE AIGC图像质量评估赛道冠军】	CVPRW NTIRE 2024	Code