数据集获取

git clone https://github.com/HCIILAB/SCUT-FBP5500-Database-Release.git

curl -L -o repo.zip https://github.com/HCIILAB/SCUT-FBP5500-Database-Release/archive/refs/heads/master.zip
unzip repo.zip

源站 README 摘录（使用方式）

SCUT-FBP5500-Database-Release

A diverse benchmark database (Size = 172MB) for multi-paradigm facial beauty prediction is now released by Human Computer Intelligent Interaction Lab of South China University of Technology. The database can be downloaded through the following links:

• Download link1 (faster for people in China):
  https://pan.baidu.com/s/1Ff2W2VLJ1ZbWSeV5JbF0Iw (PASSWORD: if7p)
• Download link2 (faster for people in other places):
  https://drive.google.com/open?id=1w0TorBfTIqbquQVd6k3h_77ypnrvfGwf

1 Description

The SCUT-FBP5500 dataset has totally 5500 frontal faces with diverse properties
(male/female, Asian/Caucasian, ages) and diverse labels (facial landmarks, beauty scores in 5 scales, beauty score distribution), which allows different computational model with different facial beauty prediction paradigms, such as appearance-based/shape-based facial beauty classification/regression/ranking model for male/female of Asian/Caucasian.

2 Database Construction

The SCUT-FBP5500 Dataset can be divided into four subsets with different races and gender, including 2000 Asian females(AF), 2000 Asian males(AM), 750 Caucasian females(CF) and 750 Caucasian males(CM). Most of the images of the SCUT-FBP5500 were collected from Internet, where some portions of Asian faces were from the DataTang, GuangZhouXiangSu and our laboratory, and some Caucasian faces were from the 10k US Adult Faces database.
All the images are labeled with beauty scores ranging from [1, 5] by totally 60 volunteers, and 86 facial landmarks are also located to the significant facial components of each images. Specifically, we save the facial landmarks in ‘pts’ format, which can be converted to ‘‘txt’’ format by running pts2txt.py. We developed several web-based GUI systems to obtain the facial beauty scores and facial landmark locations, respectively.

Training/Testing Set Split

We use two kinds of experimental settings to evaluate the facial beauty prediction methods on SCUT-FBP5500 benchmark, which includes:

1. 5-folds cross validation. For each validation, 80% samples (4400 images) are used for training and the rest (1100 images) are used for testing.
2. The split of 60% training and 40% testing. 60% samples (3300 images) are used for training and the rest (2200 images) are used for testing.
   We have provided the training and testing files in this link.

3 Train

数据加载示例（图像类）

from PIL import Image
import glob, os

files = (glob.glob(os.path.join(path, "**", "*.png"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.jpg"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.tif"), recursive=True))
print("图像文件数:", len(files))
img = Image.open(files[0]); print("尺寸/模式:", img.size, img.mode)

# torchvision Dataset 方式:
# from torchvision import datasets
# ds = datasets.ImageFolder(path)  # 要求 子目录=类别

目录组织与标注格式以源站说明和下载后实际文件为准。

完整仓库：github.com/HCIILAB/SCUT-FBP5500-Database-Release