https://github.com/routinelife1/vs-drba
vapoursynth version for DRBA
https://github.com/routinelife1/vs-drba
accelerate anime interpolation realtime rife tensorrt trt vapoursynth vfi video-frame-interpolation vs
Last synced: 2 months ago
JSON representation
vapoursynth version for DRBA
- Host: GitHub
- URL: https://github.com/routinelife1/vs-drba
- Owner: routineLife1
- License: mit
- Created: 2025-03-06T08:00:37.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2025-03-28T05:49:19.000Z (about 1 year ago)
- Last Synced: 2026-02-24T05:45:00.473Z (4 months ago)
- Topics: accelerate, anime, interpolation, realtime, rife, tensorrt, trt, vapoursynth, vfi, video-frame-interpolation, vs
- Language: Python
- Homepage:
- Size: 23.4 MB
- Stars: 13
- Watchers: 1
- Forks: 0
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# VS-DRBA
Distance Ratio Based Adjuster for Animeinterp, based on https://github.com/routineLife1/DRBA and https://github.com/HolyWu/vs-rife.
> This project is modified from [HolyWu/vs-rife](https://github.com/HolyWu/vs-rife) and achieves nearly the same interpolation quality as the original [DRBA](https://github.com/routineLife1/DRBA) project.
>
> With TensorRT integration, it achieves a 400% acceleration, enabling real-time playback on high-performance NVIDIA GPUs.
>
## Dependencies
- [PyTorch](https://pytorch.org/get-started/) 2.6.0 or later
- [VapourSynth](http://www.vapoursynth.com/) R66 or later
- [vs-miscfilters-obsolete](https://github.com/vapoursynth/vs-miscfilters-obsolete) (only needed for scene change detection)
- [cupy](https://github.com/cupy/cupy) cupy-cuda11x or later (critical for acceleration)
`trt` requires additional packages:
- [TensorRT](https://developer.nvidia.com/tensorrt) 10.7.0.post1 or later
- [Torch-TensorRT](https://pytorch.org/TensorRT/) 2.6.0 or later
To install the latest stable version of PyTorch, Torch-TensorRT and cupy, run:
```
pip install -U packaging setuptools wheel
pip install -U torch torchvision torch_tensorrt --index-url https://download.pytorch.org/whl/cu126 --extra-index-url https://pypi.nvidia.com
pip install -U cupy-cuda12x
```
## Installation
```
pip install -U vsdrba==1.0.2
```
If you want to download all models at once, run `python -m vsdrba`. If you prefer to only download the model you
specified at first run, set `auto_download=True` in `drba_rife()`.
## Usage
```python
from vsdrba import drba_rife
ret = drba_rife(clip)
```
See `__init__.py` for the description of the parameters.
## Benchmarks
| model | scale | os | hardware | arch | fps 720 | fps 1080 | vram 720 | vram 1080 | backend | verified output | batch | level | streams | threads | onnx | onnxslim / onnxsim | onnx shape | trtexec shape | precision | usage |
|----------------------| ----- | ----- |--------------------|------------------------------------------------------------|---------|----------|----------|-----------|--------------------------------------------------------------------------------| ---------------------------------- | ----- | ----- |---------|---------| --------- | ------------------ | ----------- | ------------- | --------- |-----------------------------------------------------------------------------------------------------|
| rife 4.26 heavy | 2x | Linux | 3070laptop / 12400 | [rife](https://github.com/hzwer/Practical-RIFE) (4.26) | 119 | 53 | 1.6gb | 3.4gb | trt 10.8, torch 20241231+cu126, torch_trt 20250102+cu126 (holywu vsrife) | yes, works | 1 | 5 | - | 8 | - | - | - | static | RGBH | rife(clip, trt=True, trt_static_shape=True, model="4.26.heavy", trt_optimization_level=5, sc=False) |
| drba_rife 4.26 heavy | 2x | Linux | 3070laptop / 12400 | [drba_rife](https://github.com/routineLife1/DRBA) (4.26) | 158 | 70 | 1.7gb | 3.7gb | trt 10.8, torch 20241231+cu126, torch_trt 20250102+cu126 (routineLife1 vsdrba) | yes, works | 1 | 5 | - | 8 | - | - | - | static | RGBH | rife(clip, trt=True, trt_static_shape=True, model="4.26.heavy", trt_optimization_level=5, sc=False) |