---
tags:
- pytorch
- vision
- face-analysis
- beauty-prediction
- scut-fbp5500
library_name: generic
license: mit
---

# SCUT-FBP5500 ResNet-18 Model

This repository contains the PyTorch weights for a facial beauty prediction model trained on the [SCUT-FBP5500 dataset](https://github.com/HCIILAB/SCUT-FBP5500-Database-Release).

## Model Files

- **`resnet18.pth`**: Original weights (saved in Python 2). Requires `encoding='latin1'` to load in Python 3.
- **`resnet18_py3.pth`**: Converted weights compatible with Python 3+ (Recommended).

## Model Details

- **Architecture:** ResNet-18
- **Dataset:** SCUT-FBP5500 (Facial Beauty Prediction)
- **Framework:** PyTorch

## Origin

The original weights were provided by [brian-xu](https://github.com/brian-xu/torch-converter). The `resnet18_py3.pth` file is a converted version using `_use_new_zipfile_serialization=True`.

## Usage (Python 3)

```python
import torch
import torchvision.models as models

model = models.resnet18(num_classes=1) 

# Load the converted weights (Recommended)
state_dict = torch.load("resnet18_py3.pth")
model.load_state_dict(state_dict)
model.eval()
```

## Licensing

Please refer to the original [SCUT-FBP5500 Database](https://github.com/HCIILAB/SCUT-FBP5500-Database-Release) for usage terms regarding the dataset and derived models.