Real-ESRGAN-TPU v2.0

Run Real-ESRGAN on Sophon TPU

chip support: SG2300X, BM1684X

platform support: ARM, X86

Tested device: Radxa Fogwise® AirBox, Sophon SC7+

✨Support run by gradio or streamlit (do not maintain)

🌟Support upscale video and image

🌠Support models: RealESRGAN_x4plus, RealESRGAN_x4plus_anime_6B, realesr-animevideo_v3, realesr-general-x4v3

🏆Support Face Enhance: CodeFormer, GFPGAN

🌟Support Background Remove: RMBG

🌠Support DIY Background Picture

Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data

[Paper]   [YouTube Video]   [B站讲解]   [Poster]   [PPT slides] Xintao Wang, Liangbin Xie, Chao Dong, Ying Shan Tencent ARC Lab; Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences

Usage

• Clone the repo

  git clone https://github.com/zifeng-radxa/Real-ESRGAN-tpu_Web.git

• Clone the tool-box

  cd Real-ESRGAN-tpu_Web
  git clone https://github.com/zifeng-radxa/FACEXLIB.git

• Setup environment

  pip3 install -r requirements.txt
  ## aarch64
  pip3 install https://github.com/radxa-edge/TPU-Edge-AI/releases/download/v0.1.0/tpu_perf-1.2.31-py3-none-manylinux2014_aarch64.whl
  ## x86
  pip3 install https://github.com/radxa-edge/TPU-Edge-AI/releases/download/v0.1.0/tpu_perf-1.2.31-py3-none-manylinux2014_x86_64.whl

• Instal FFmpeg (do not support sophon-ffmpeg)

  sudo apt install ffmpeg

• Download models

  python3 download_models.py

• Run the boot script

  • Run by gradio

  bash run_gr.sh

  - (Optional) Run by streamlit

  bash run_st.sh