TFDAT

TFDAT (Temporal FDAT) is a custom VSR architecture. This is a continuation of my previous VSR architectures, namely TSPANv2. TFDAT is a major step up in every way, providing significantly better quality, temporal coherency, and even inference speed! This arch has support for PyTorch, ONNX, and TensorRT!

This repository only hosts the Pytorch inference code. To train a TFDAT model, you'll want to use traiNNer-redux with the TFDAT config and a video dataset.

For easier inference than the GUI provided in this codebase, try out Vapourkit. To make a video dataset, try my other tool, video destroyer.

Getting Started

1. Clone the repository:

   git clone https://github.com/Kim2091/TFDAT
   

2. Install PyTorch with CUDA: Follow the instructions at pytorch.org.

3. Install required packages:

   pip install -r requirements.txt
   

Usage

You can use TFDAT through Vapourkit (preferred), the included GUI (directions below), or the command line.

TensorRT

For high-performance inference, refer to the TensorRT guide.

GUI Usage

For an easy-to-use experience with PyTorch or ONNX models, launch the GUI:

python vsr_gui.py

Command-Line Usage

For more advanced control, you can use the command-line scripts.

Video upscaling (PyTorch):

python test_vsr.py --model_path pretrained_models/tfdat.pth --input path/to/video.mp4 --output path/to/output.mp4

Key arguments for test_vsr.py and test_onnx.py:

• --video_codec: Specify the video codec (e.g., libx264, libx265).
• --crf: Set the Constant Rate Factor for quality (for libx264/libx265).

ONNX Conversion

Unlike my previous repositories, ONNX conversion is now done within traiNNer-redux instead. Follow these directions: https://trainner-redux.readthedocs.io/en/latest/getting_started.html#convert-models-to-onnx

Credits (thanks all!)

Thank you to leobby and Bendel for testing the arch!

• Folder structure and video processing code is derived from SCUNet

• The TFDAT architecture is based on FDAT with extensive modifications