|
|
--- |
|
|
license: mit |
|
|
library_name: pytorch |
|
|
tags: |
|
|
- video-enhancement |
|
|
- super-resolution |
|
|
- frame-interpolation |
|
|
- computer-vision |
|
|
pipeline_tag: image-to-image |
|
|
--- |
|
|
|
|
|
# Nightingale Video Enhancement Models |
|
|
|
|
|
This repository contains the pre-trained AI models used by the Nightingale Video Enhancement Service. |
|
|
|
|
|
## Models Included |
|
|
|
|
|
### Real-ESRGAN (Super Resolution) |
|
|
- **File**: `RealESRGAN_x4plus.pth` |
|
|
- **Purpose**: Upscales video resolution (e.g., 480p → 1080p) |
|
|
- **Architecture**: Real-ESRGAN x4 plus model |
|
|
- **License**: BSD 3-Clause |
|
|
- **Source**: [Real-ESRGAN Repository](https://github.com/xinntao/Real-ESRGAN) |
|
|
|
|
|
### RIFE (Frame Interpolation) |
|
|
- **File**: `RIFE_v4.6.pth` |
|
|
- **Purpose**: Interpolates frames to increase frame rate (e.g., 15fps → 30fps) |
|
|
- **Architecture**: RIFE v4.6 model |
|
|
- **License**: MIT License |
|
|
- **Source**: [RIFE Repository](https://github.com/megvii-research/ECCV2022-RIFE) |
|
|
|
|
|
## Usage |
|
|
|
|
|
These models are automatically downloaded by the Nightingale Video Enhancement Service. To use them in your own project: |
|
|
|
|
|
```python |
|
|
from huggingface_hub import hf_hub_download |
|
|
|
|
|
# Download Real-ESRGAN model |
|
|
realesrgan_path = hf_hub_download( |
|
|
repo_id="mkulyma/nightingale-models", |
|
|
filename="RealESRGAN_x4plus.pth" |
|
|
) |
|
|
|
|
|
# Download RIFE model |
|
|
rife_path = hf_hub_download( |
|
|
repo_id="mkulyma/nightingale-models", |
|
|
filename="RIFE_v4.6.pth" |
|
|
) |
|
|
``` |
|
|
|
|
|
## Model Details |
|
|
|
|
|
- **Total Size**: ~150MB |
|
|
- **Framework**: PyTorch |
|
|
- **Device Support**: CUDA GPU (recommended) or CPU |
|
|
- **Input**: Video files (MP4, AVI, MOV, etc.) |
|
|
- **Output**: Enhanced video with higher resolution and/or frame rate |
|
|
|
|
|
## Nightingale Project |
|
|
|
|
|
These models are part of the [Nightingale Video Enhancement Service](https://github.com/your-username/nightingale), an AI-powered video enhancement platform that provides: |
|
|
|
|
|
- Web-based interface for video upload and processing |
|
|
- Docker deployment for easy setup |
|
|
- GPU acceleration with CPU fallback |
|
|
- Real-time processing monitoring |
|
|
- Batch processing capabilities |
|
|
|
|
|
## License |
|
|
|
|
|
The models retain their original licenses: |
|
|
- Real-ESRGAN: BSD 3-Clause License |
|
|
- RIFE: MIT License |
|
|
|
|
|
This model repository packaging: MIT License |
|
|
|
