Upload 415 files

.gitattributes

@@ -33,3 +33,11 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+checkpoints/pretrained.state filter=lfs diff=lfs merge=lfs -text
+examples/kennedy_hd.mkv filter=lfs diff=lfs merge=lfs -text
+examples/mona_hd.jpg filter=lfs diff=lfs merge=lfs -text
+examples/mona_hd.mkv filter=lfs diff=lfs merge=lfs -text
+input_audios/part_000_RVC_1.wav filter=lfs diff=lfs merge=lfs -text
+input_videos/MC.mp4 filter=lfs diff=lfs merge=lfs -text
+temp/result.avi filter=lfs diff=lfs merge=lfs -text
+temp/temp.wav filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,121 @@
+# Wav2Lip-HD: Improving Wav2Lip to achieve High-Fidelity Videos
+
+This repository contains code for achieving high-fidelity lip-syncing in videos, using the [Wav2Lip algorithm](https://github.com/Rudrabha/Wav2Lip) for lip-syncing and the [Real-ESRGAN algorithm](https://github.com/xinntao/Real-ESRGAN) for super-resolution. The combination of these two algorithms allows for the creation of lip-synced videos that are both highly accurate and visually stunning.
+
+## Algorithm
+
+The algorithm for achieving high-fidelity lip-syncing with Wav2Lip and Real-ESRGAN can be summarized as follows:
+
+1. The input video and audio are given to `Wav2Lip` algorithm.
+2. Python script is written to extract frames from the video generated by wav2lip.
+3. Frames are provided to Real-ESRGAN algorithm to improve quality.
+4. Then, the high-quality frames are converted to video using ffmpeg, along with the original audio.
+5. The result is a high-quality lip-syncing video.
+6. The specific steps for running this algorithm are described in the [Testing Model](https://github.com/saifhassan/Wav2Lip-HD#testing-model) section of this README.
+
+## Testing Model
+
+To test the "Wav2Lip-HD" model, follow these steps:
+
+1. Clone this repository and install requirements using following command (Make sure, Python and CUDA are already installed):
+
+    ```
+    git clone https://github.com/saifhassan/Wav2Lip-HD.git
+    cd Wav2Lip-HD
+    pip install -r requirements.txt
+    ```
+    
+2. Downloading weights
+
+| Model        | Directory           | Download Link  |
+| :------------- |:-------------| :-----:|
+| Wav2Lip           | [checkpoints/](https://github.com/saifhassan/Wav2Lip-HD/tree/main/checkpoints)   | [Link](https://drive.google.com/drive/folders/1tB_uz-TYMePRMZzrDMdShWUZZ0JK3SIZ?usp=sharing) |
+| ESRGAN            | [experiments/001_ESRGAN_x4_f64b23_custom16k_500k_B16G1_wandb/models/](https://github.com/saifhassan/Wav2Lip-HD/tree/main/experiments/001_ESRGAN_x4_f64b23_custom16k_500k_B16G1_wandb/models) | [Link](https://drive.google.com/file/d/1Al8lEpnx2K-kDX7zL2DBcAuDnSKXACPb/view?usp=sharing) |
+| Face_Detection    | [face_detection/detection/sfd/](https://github.com/saifhassan/Wav2Lip-HD/tree/main/face_detection/detection/sfd) | [Link](https://drive.google.com/file/d/1uNLYCPFFmO-og3WSHyFytJQLLYOwH5uY/view?usp=sharing) |
+| Real-ESRGAN       | Real-ESRGAN/gfpgan/weights/   | [Link](https://drive.google.com/drive/folders/1BLx6aMpHgFt41fJ27_cRmT8bt53kVAYG?usp=sharing) |
+| Real-ESRGAN       | Real-ESRGAN/weights/          | [Link](https://drive.google.com/file/d/1qNIf8cJl_dQo3ivelPJVWFkApyEAGnLi/view?usp=sharing) |
+
+
+3. Put input video to `input_videos` directory and input audio to `input_audios` directory.
+4. Open `run_final.sh` file and modify following parameters:
+ 
+     `filename=kennedy` (just video file name without extension)
+     
+     `input_audio=input_audios/ai.wav` (audio filename with extension)
+
+5. Execute `run_final.sh` using following command:
+
+    ```
+    bash run_final.sh
+    ```
+    
+6. Outputs
+
+- `output_videos_wav2lip` directory contains video output generated by wav2lip algorithm.
+- `frames_wav2lip` directory contains frames extracted from video (generated by wav2lip algorithm).
+- `frames_hd` directory contains frames after performing super-resolution using Real-ESRGAN algorithm.
+- `output_videos_hd` directory contains final high quality video output generated by Wav2Lip-HD.
+
+
+## Results
+The results produced by Wav2Lip-HD are in two forms, one is frames and other is videos. Both are shared below:
+
+### Example output frames </summary>
+<table>
+  <tr>
+    <td>Frame by Wav2Lip</td>
+     <td>Optimized Frame</td>
+  </tr>
+  <tr>
+    <td><img src="examples/1_low.jpg" width=500></td>
+    <td><img src="examples/1_hd.jpg" width=500></td>
+  </tr>
+    <tr>
+    <td><img src="examples/kennedy_low.jpg" width=500></td>
+    <td><img src="examples/kennedy_hd.jpg" width=500></td>
+  </tr>
+
+  </tr>
+    <tr>
+    <td><img src="examples/mona_low.jpg" width=500></td>
+    <td><img src="examples/mona_hd.jpg" width=500></td>
+  </tr>
+ </table>
+ </Details>
+
+ ### Example output videos
+    
+ | Video by Wav2Lip  | Optimized Video |
+| ------------- | ------------- |
+| <video src="https://user-images.githubusercontent.com/11873763/229389410-56d96244-8c67-4add-a43e-a4900aa9db88.mp4" width="500">  | <video src="https://user-images.githubusercontent.com/11873763/229389414-d5cb6d33-7772-47a7-b829-9e3d5c3945a1.mp4" width="500">|
+| <video src="https://user-images.githubusercontent.com/11873763/229389751-507669f1-7772-4863-ab23-8df7f206a065.mp4" width="500">  | <video src="https://user-images.githubusercontent.com/11873763/229389962-5373b765-ce3a-4af2-bd6a-8be8543ee933.mp4" width="500">|
+
+## Acknowledgements
+
+We would like to thank the following repositories and libraries for their contributions to our work:
+
+1. The [Wav2Lip](https://github.com/Rudrabha/Wav2Lip) repository, which is the core model of our algorithm that performs lip-sync.
+2. The [face-parsing.PyTorch](https://github.com/zllrunning/face-parsing.PyTorch) repository, which provides us with a model for face segmentation.
+3. The [Real-ESRGAN](https://github.com/xinntao/Real-ESRGAN) repository, which provides the super resolution component for our algorithm.
+4. [ffmpeg](https://ffmpeg.org), which we use for converting frames to video.
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+    
+
+

Real-ESRGAN/.pre-commit-config.yaml

@@ -0,0 +1,46 @@
+repos:
+  # flake8
+  - repo: https://github.com/PyCQA/flake8
+    rev: 3.8.3
+    hooks:
+      - id: flake8
+        args: ["--config=setup.cfg", "--ignore=W504, W503"]
+
+  # modify known_third_party
+  - repo: https://github.com/asottile/seed-isort-config
+    rev: v2.2.0
+    hooks:
+      - id: seed-isort-config
+
+  # isort
+  - repo: https://github.com/timothycrosley/isort
+    rev: 5.2.2
+    hooks:
+      - id: isort
+
+  # yapf
+  - repo: https://github.com/pre-commit/mirrors-yapf
+    rev: v0.30.0
+    hooks:
+      - id: yapf
+
+  # codespell
+  - repo: https://github.com/codespell-project/codespell
+    rev: v2.1.0
+    hooks:
+      - id: codespell
+
+  # pre-commit-hooks
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v3.2.0
+    hooks:
+      - id: trailing-whitespace  # Trim trailing whitespace
+      - id: check-yaml  # Attempt to load all yaml files to verify syntax
+      - id: check-merge-conflict  # Check for files that contain merge conflict strings
+      - id: double-quote-string-fixer  # Replace double quoted strings with single quoted strings
+      - id: end-of-file-fixer  # Make sure files end in a newline and only a newline
+      - id: requirements-txt-fixer  # Sort entries in requirements.txt and remove incorrect entry for pkg-resources==0.0.0
+      - id: fix-encoding-pragma  # Remove the coding pragma: # -*- coding: utf-8 -*-
+        args: ["--remove"]
+      - id: mixed-line-ending  # Replace or check mixed line ending
+        args: ["--fix=lf"]

Real-ESRGAN/MANIFEST.in

@@ -0,0 +1,8 @@
+include assets/*
+include inputs/*
+include scripts/*.py
+include inference_realesrgan.py
+include VERSION
+include LICENSE
+include requirements.txt
+include weights/README.md

Real-ESRGAN/README_CN.md

@@ -0,0 +1,276 @@
+<p align="center">
+  <img src="assets/realesrgan_logo.png" height=120>
+</p>
+
+## <div align="center"><b><a href="README.md">English</a> | <a href="README_CN.md">简体中文</a></b></div>
+
+[![download](https://img.shields.io/github/downloads/xinntao/Real-ESRGAN/total.svg)](https://github.com/xinntao/Real-ESRGAN/releases)
+[![PyPI](https://img.shields.io/pypi/v/realesrgan)](https://pypi.org/project/realesrgan/)
+[![Open issue](https://img.shields.io/github/issues/xinntao/Real-ESRGAN)](https://github.com/xinntao/Real-ESRGAN/issues)
+[![Closed issue](https://img.shields.io/github/issues-closed/xinntao/Real-ESRGAN)](https://github.com/xinntao/Real-ESRGAN/issues)
+[![LICENSE](https://img.shields.io/github/license/xinntao/Real-ESRGAN.svg)](https://github.com/xinntao/Real-ESRGAN/blob/master/LICENSE)
+[![python lint](https://github.com/xinntao/Real-ESRGAN/actions/workflows/pylint.yml/badge.svg)](https://github.com/xinntao/Real-ESRGAN/blob/master/.github/workflows/pylint.yml)
+[![Publish-pip](https://github.com/xinntao/Real-ESRGAN/actions/workflows/publish-pip.yml/badge.svg)](https://github.com/xinntao/Real-ESRGAN/blob/master/.github/workflows/publish-pip.yml)
+
+:fire: 更新动漫视频的小模型 **RealESRGAN AnimeVideo-v3**. 更多信息在 [[动漫视频模型介绍](docs/anime_video_model.md)] 和 [[比较](docs/anime_comparisons_CN.md)] 中.
+
+1. Real-ESRGAN的[Colab Demo](https://colab.research.google.com/drive/1k2Zod6kSHEvraybHl50Lys0LerhyTMCo?usp=sharing) | Real-ESRGAN**动漫视频** 的[Colab Demo](https://colab.research.google.com/drive/1yNl9ORUxxlL4N0keJa2SEPB61imPQd1B?usp=sharing)
+2. **支持Intel/AMD/Nvidia显卡**的绿色版exe文件： [Windows版](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesrgan-ncnn-vulkan-20220424-windows.zip) / [Linux版](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesrgan-ncnn-vulkan-20220424-ubuntu.zip) / [macOS版](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesrgan-ncnn-vulkan-20220424-macos.zip)，详情请移步[这里](#便携版（绿色版）可执行文件)。NCNN的实现在 [Real-ESRGAN-ncnn-vulkan](https://github.com/xinntao/Real-ESRGAN-ncnn-vulkan)。
+
+Real-ESRGAN 的目标是开发出**实用的图像/视频修复算法**。<br>
+我们在 ESRGAN 的基础上使用纯合成的数据来进行训练，以使其能被应用于实际的图片修复的场景（顾名思义：Real-ESRGAN）。
+
+:art: Real-ESRGAN 需要，也很欢迎你的贡献，如新功能、模型、bug修复、建议、维护等等。详情可以查看[CONTRIBUTING.md](docs/CONTRIBUTING.md)，所有的贡献者都会被列在[此处](README_CN.md#hugs-感谢)。
+
+:milky_way: 感谢大家提供了很好的反馈。这些反馈会逐步更新在 [这个文档](docs/feedback.md)。
+
+:question: 常见的问题可以在[FAQ.md](docs/FAQ.md)中找到答案。（好吧，现在还是空白的=-=||）
+
+---
+
+如果 Real-ESRGAN 对你有帮助，可以给本项目一个 Star :star: ，或者推荐给你的朋友们，谢谢！:blush: <br/>
+其他推荐的项目：<br/>
+:arrow_forward: [GFPGAN](https://github.com/TencentARC/GFPGAN): 实用的人脸复原算法 <br>
+:arrow_forward: [BasicSR](https://github.com/xinntao/BasicSR): 开源的图像和视频工具箱<br>
+:arrow_forward: [facexlib](https://github.com/xinntao/facexlib): 提供与人脸相关的工具箱<br>
+:arrow_forward: [HandyView](https://github.com/xinntao/HandyView): 基于PyQt5的图片查看器，方便查看以及比较 <br>
+
+---
+
+<!---------------------------------- Updates --------------------------->
+<details>
+<summary>🚩<b>更新</b></summary>
+
+- ✅ 更新动漫视频的小模型 **RealESRGAN AnimeVideo-v3**. 更多信息在 [anime video models](docs/anime_video_model.md) 和 [comparisons](docs/anime_comparisons.md)中.
+- ✅ 添加了针对动漫视频的小模型, 更多信息在 [anime video models](docs/anime_video_model.md) 中.
+- ✅ 添加了ncnn 实现：[Real-ESRGAN-ncnn-vulkan](https://github.com/xinntao/Real-ESRGAN-ncnn-vulkan).
+- ✅ 添加了 [*RealESRGAN_x4plus_anime_6B.pth*](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth)，对二次元图片进行了优化，并减少了model的大小。详情 以及 与[waifu2x](https://github.com/nihui/waifu2x-ncnn-vulkan)的对比请查看[**anime_model.md**](docs/anime_model.md)
+- ✅支持用户在自己的数据上进行微调 (finetune)：[详情](docs/Training.md#Finetune-Real-ESRGAN-on-your-own-dataset)
+- ✅ 支持使用[GFPGAN](https://github.com/TencentARC/GFPGAN)**增强人脸**
+- ✅ 通过[Gradio](https://github.com/gradio-app/gradio)添加到了[Huggingface Spaces](https://huggingface.co/spaces)（一个机器学习应用的在线平台）：[Gradio在线版](https://huggingface.co/spaces/akhaliq/Real-ESRGAN)。感谢[@AK391](https://github.com/AK391)
+- ✅ 支持任意比例的缩放：`--outscale`（实际上使用`LANCZOS4`来更进一步调整输出图像的尺寸）。添加了*RealESRGAN_x2plus.pth*模型
+- ✅ [推断脚本](inference_realesrgan.py)支持: 1) 分块处理**tile**; 2) 带**alpha通道**的图像; 3) **灰色**图像; 4) **16-bit**图像.
+- ✅ 训练代码已经发布，具体做法可查看：[Training.md](docs/Training.md)。
+
+</details>
+
+<!---------------------------------- Projects that use Real-ESRGAN --------------------------->
+<details>
+<summary>🧩<b>使用Real-ESRGAN的项目</b></summary>
+
+&nbsp;&nbsp;&nbsp;&nbsp;👋 如果你开发/使用/集成了Real-ESRGAN, 欢迎联系我添加
+
+- NCNN-Android: [RealSR-NCNN-Android](https://github.com/tumuyan/RealSR-NCNN-Android) by [tumuyan](https://github.com/tumuyan)
+- VapourSynth: [vs-realesrgan](https://github.com/HolyWu/vs-realesrgan) by [HolyWu](https://github.com/HolyWu)
+- NCNN: [Real-ESRGAN-ncnn-vulkan](https://github.com/xinntao/Real-ESRGAN-ncnn-vulkan)
+
+&nbsp;&nbsp;&nbsp;&nbsp;**易用的图形界面**
+
+- [Waifu2x-Extension-GUI](https://github.com/AaronFeng753/Waifu2x-Extension-GUI) by [AaronFeng753](https://github.com/AaronFeng753)
+- [Squirrel-RIFE](https://github.com/Justin62628/Squirrel-RIFE) by [Justin62628](https://github.com/Justin62628)
+- [Real-GUI](https://github.com/scifx/Real-GUI) by [scifx](https://github.com/scifx)
+- [Real-ESRGAN_GUI](https://github.com/net2cn/Real-ESRGAN_GUI) by [net2cn](https://github.com/net2cn)
+- [Real-ESRGAN-EGUI](https://github.com/WGzeyu/Real-ESRGAN-EGUI) by [WGzeyu](https://github.com/WGzeyu)
+- [anime_upscaler](https://github.com/shangar21/anime_upscaler) by [shangar21](https://github.com/shangar21)
+- [RealESRGAN-GUI](https://github.com/Baiyuetribe/paper2gui/blob/main/Video%20Super%20Resolution/RealESRGAN-GUI.md) by [Baiyuetribe](https://github.com/Baiyuetribe)
+
+</details>
+
+<details>
+<summary>👀<b>Demo视频（B站）</b></summary>
+
+- [大闹天宫片段](https://www.bilibili.com/video/BV1ja41117zb)
+
+</details>
+
+### :book: Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data
+
+> [[论文](https://arxiv.org/abs/2107.10833)] &emsp; [项目主页] &emsp; [[YouTube 视频](https://www.youtube.com/watch?v=fxHWoDSSvSc)] &emsp; [[B站视频](https://www.bilibili.com/video/BV1H34y1m7sS/)] &emsp; [[Poster](https://xinntao.github.io/projects/RealESRGAN_src/RealESRGAN_poster.pdf)] &emsp; [[PPT](https://docs.google.com/presentation/d/1QtW6Iy8rm8rGLsJ0Ldti6kP-7Qyzy6XL/edit?usp=sharing&ouid=109799856763657548160&rtpof=true&sd=true)]<br>
+> [Xintao Wang](https://xinntao.github.io/), Liangbin Xie, [Chao Dong](https://scholar.google.com.hk/citations?user=OSDCB0UAAAAJ), [Ying Shan](https://scholar.google.com/citations?user=4oXBp9UAAAAJ&hl=en) <br>
+> Tencent ARC Lab; Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences
+
+<p align="center">
+  <img src="assets/teaser.jpg">
+</p>
+
+---
+
+我们提供了一套训练好的模型（*RealESRGAN_x4plus.pth*)，可以进行4倍的超分辨率。<br>
+**现在的 Real-ESRGAN 还是有几率失败的，因为现实生活的降质过程比较复杂。**<br>
+而且，本项目对**人脸以及文字之类**的效果还不是太好，但是我们会持续进行优化的。<br>
+
+Real-ESRGAN 将会被长期支持，我会在空闲的时间中持续维护更新。
+
+这些是未来计划的几个新功能：
+
+- [ ] 优化人脸
+- [ ] 优化文字
+- [x] 优化动画图像
+- [ ] 支持更多的超分辨率比例
+- [ ] 可调节的复原
+
+如果你有好主意或需求，欢迎在 issue 或 discussion 中提出。<br/>
+如果你有一些 Real-ESRGAN 中有问题的照片，你也可以在 issue 或者 discussion 中发出来。我会留意（但是不一定能解决:stuck_out_tongue:）。如果有必要的话，我还会专门开一页来记录那些有待解决的图像。
+
+---
+
+### 便携版（绿色版）可执行文件
+
+你可以下载**支持Intel/AMD/Nvidia显卡**的绿色版exe文件： [Windows版](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesrgan-ncnn-vulkan-20220424-windows.zip) / [Linux版](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesrgan-ncnn-vulkan-20220424-ubuntu.zip) / [macOS版](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesrgan-ncnn-vulkan-20220424-macos.zip)。
+
+绿色版指的是这些exe你可以直接运行（放U盘里拷走都没问题），因为里面已经有所需的文件和模型了。它不需要 CUDA 或者 PyTorch运行环境。<br>
+
+你可以通过下面这个命令来运行（Windows版本的例子，更多信息请查看对应版本的README.md）：
+
+```bash
+./realesrgan-ncnn-vulkan.exe -i 输入图像.jpg -o 输出图像.png -n 模型名字
+```
+
+我们提供了五种模型：
+
+1. realesrgan-x4plus（默认）
+2. reaesrnet-x4plus
+3. realesrgan-x4plus-anime（针对动漫插画图像优化，有更小的体积）
+4. realesr-animevideov3 (针对动漫视频)
+
+你可以通过`-n`参数来使用其他模型，例如`./realesrgan-ncnn-vulkan.exe -i 二次元图片.jpg -o 二刺螈图片.png -n realesrgan-x4plus-anime`
+
+### 可执行文件的用法
+
+1. 更多细节可以参考 [Real-ESRGAN-ncnn-vulkan](https://github.com/xinntao/Real-ESRGAN-ncnn-vulkan#computer-usages).
+2. 注意：可执行文件并没有支持 python 脚本 `inference_realesrgan.py` 中所有的功能，比如 `outscale` 选项) .
+
+```console
+Usage: realesrgan-ncnn-vulkan.exe -i infile -o outfile [options]...
+
+  -h                   show this help
+  -i input-path        input image path (jpg/png/webp) or directory
+  -o output-path       output image path (jpg/png/webp) or directory
+  -s scale             upscale ratio (can be 2, 3, 4. default=4)
+  -t tile-size         tile size (>=32/0=auto, default=0) can be 0,0,0 for multi-gpu
+  -m model-path        folder path to the pre-trained models. default=models
+  -n model-name        model name (default=realesr-animevideov3, can be realesr-animevideov3 | realesrgan-x4plus | realesrgan-x4plus-anime | realesrnet-x4plus)
+  -g gpu-id            gpu device to use (default=auto) can be 0,1,2 for multi-gpu
+  -j load:proc:save    thread count for load/proc/save (default=1:2:2) can be 1:2,2,2:2 for multi-gpu
+  -x                   enable tta mode"
+  -f format            output image format (jpg/png/webp, default=ext/png)
+  -v                   verbose output
+```
+
+由于这些exe文件会把图像分成几个板块，然后来分别进行处理，再合成导出，输出的图像可能会有一点割裂感（而且可能跟PyTorch的输出不太一样）
+
+---
+
+## :wrench: 依赖以及安装
+
+- Python >= 3.7 (推荐使用[Anaconda](https://www.anaconda.com/download/#linux)或[Miniconda](https://docs.conda.io/en/latest/miniconda.html))
+- [PyTorch >= 1.7](https://pytorch.org/)
+
+#### 安装
+
+1. 把项目克隆到本地
+
+    ```bash
+    git clone https://github.com/xinntao/Real-ESRGAN.git
+    cd Real-ESRGAN
+    ```
+
+2. 安装各种依赖
+
+    ```bash
+    # 安装 basicsr - https://github.com/xinntao/BasicSR
+    # 我们使用BasicSR来训练以及推断
+    pip install basicsr
+    # facexlib和gfpgan是用来增强人脸的
+    pip install facexlib
+    pip install gfpgan
+    pip install -r requirements.txt
+    python setup.py develop
+    ```
+
+## :zap: 快速上手
+
+### 普通图片
+
+下载我们训练好的模型: [RealESRGAN_x4plus.pth](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth)
+
+```bash
+wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth -P weights
+```
+
+推断!
+
+```bash
+python inference_realesrgan.py -n RealESRGAN_x4plus -i inputs --face_enhance
+```
+
+结果在`results`文件夹
+
+### 动画图片
+
+<p align="center">
+  <img src="https://raw.githubusercontent.com/xinntao/public-figures/master/Real-ESRGAN/cmp_realesrgan_anime_1.png">
+</p>
+
+训练好的模型: [RealESRGAN_x4plus_anime_6B](https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth)<br>
+有关[waifu2x](https://github.com/nihui/waifu2x-ncnn-vulkan)的更多信息和对比在[**anime_model.md**](docs/anime_model.md)中。
+
+```bash
+# 下载模型
+wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth -P weights
+# 推断
+python inference_realesrgan.py -n RealESRGAN_x4plus_anime_6B -i inputs
+```
+
+结果在`results`文件夹
+
+### Python 脚本的用法
+
+1. 虽然你使用了 X4 模型，但是你可以 **输出任意尺寸比例的图片**，只要实用了 `outscale` 参数. 程序会进一步对模型的输出图像进行缩放。
+
+```console
+Usage: python inference_realesrgan.py -n RealESRGAN_x4plus -i infile -o outfile [options]...
+
+A common command: python inference_realesrgan.py -n RealESRGAN_x4plus -i infile --outscale 3.5 --face_enhance
+
+  -h                   show this help
+  -i --input           Input image or folder. Default: inputs
+  -o --output          Output folder. Default: results
+  -n --model_name      Model name. Default: RealESRGAN_x4plus
+  -s, --outscale       The final upsampling scale of the image. Default: 4
+  --suffix             Suffix of the restored image. Default: out
+  -t, --tile           Tile size, 0 for no tile during testing. Default: 0
+  --face_enhance       Whether to use GFPGAN to enhance face. Default: False
+  --fp32               Whether to use half precision during inference. Default: False
+  --ext                Image extension. Options: auto | jpg | png, auto means using the same extension as inputs. Default: auto
+```
+
+## :european_castle: 模型库
+
+请参见 [docs/model_zoo.md](docs/model_zoo.md)
+
+## :computer: 训练，在你的数据上微调（Fine-tune）
+
+这里有一份详细的指南：[Training.md](docs/Training.md).
+
+## BibTeX 引用
+
+    @Article{wang2021realesrgan,
+        title={Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data},
+        author={Xintao Wang and Liangbin Xie and Chao Dong and Ying Shan},
+        journal={arXiv:2107.10833},
+        year={2021}
+    }
+
+## :e-mail: 联系我们
+
+如果你有任何问题，请通过 `xintao.wang@outlook.com` 或 `xintaowang@tencent.com` 联系我们。
+
+## :hugs: 感谢
+
+感谢所有的贡献者大大们~
+
+- [AK391](https://github.com/AK391): 通过[Gradio](https://github.com/gradio-app/gradio)添加到了[Huggingface Spaces](https://huggingface.co/spaces)（一个机器学习应用的在线平台）：[Gradio在线版](https://huggingface.co/spaces/akhaliq/Real-ESRGAN)。
+- [Asiimoviet](https://github.com/Asiimoviet): 把 README.md 文档 翻译成了中文。
+- [2ji3150](https://github.com/2ji3150): 感谢详尽并且富有价值的[反馈、建议](https://github.com/xinntao/Real-ESRGAN/issues/131).
+- [Jared-02](https://github.com/Jared-02): 把 Training.md 文档 翻译成了中文。

Real-ESRGAN/cog.yaml

@@ -0,0 +1,22 @@
+# This file is used for constructing replicate env
+image: "r8.im/tencentarc/realesrgan"
+
+build:
+  gpu: true
+  python_version: "3.8"
+  system_packages:
+    - "libgl1-mesa-glx"
+    - "libglib2.0-0"
+  python_packages:
+    - "torch==1.7.1"
+    - "torchvision==0.8.2"
+    - "numpy==1.21.1"
+    - "lmdb==1.2.1"
+    - "opencv-python==4.5.3.56"
+    - "PyYAML==5.4.1"
+    - "tqdm==4.62.2"
+    - "yapf==0.31.0"
+    - "basicsr==1.4.2"
+    - "facexlib==0.2.5"
+
+predict: "cog_predict.py:Predictor"

Real-ESRGAN/cog_predict.py

@@ -0,0 +1,148 @@
+# flake8: noqa
+# This file is used for deploying replicate models
+# running: cog predict -i img=@inputs/00017_gray.png -i version='General - v3' -i scale=2 -i face_enhance=True -i tile=0
+# push: cog push r8.im/xinntao/realesrgan
+
+import os
+
+os.system('pip install gfpgan')
+os.system('python setup.py develop')
+
+import cv2
+import shutil
+import tempfile
+import torch
+from basicsr.archs.rrdbnet_arch import RRDBNet
+from basicsr.archs.srvgg_arch import SRVGGNetCompact
+
+from realesrgan.utils import RealESRGANer
+
+try:
+    from cog import BasePredictor, Input, Path
+    from gfpgan import GFPGANer
+except Exception:
+    print('please install cog and realesrgan package')
+
+
+class Predictor(BasePredictor):
+
+    def setup(self):
+        os.makedirs('output', exist_ok=True)
+        # download weights
+        if not os.path.exists('weights/realesr-general-x4v3.pth'):
+            os.system(
+                'wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth -P ./weights'
+            )
+        if not os.path.exists('weights/GFPGANv1.4.pth'):
+            os.system('wget https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.4.pth -P ./weights')
+        if not os.path.exists('weights/RealESRGAN_x4plus.pth'):
+            os.system(
+                'wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth -P ./weights'
+            )
+        if not os.path.exists('weights/RealESRGAN_x4plus_anime_6B.pth'):
+            os.system(
+                'wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth -P ./weights'
+            )
+        if not os.path.exists('weights/realesr-animevideov3.pth'):
+            os.system(
+                'wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-animevideov3.pth -P ./weights'
+            )
+
+    def choose_model(self, scale, version, tile=0):
+        half = True if torch.cuda.is_available() else False
+        if version == 'General - RealESRGANplus':
+            model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=4)
+            model_path = 'weights/RealESRGAN_x4plus.pth'
+            self.upsampler = RealESRGANer(
+                scale=4, model_path=model_path, model=model, tile=tile, tile_pad=10, pre_pad=0, half=half)
+        elif version == 'General - v3':
+            model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=32, upscale=4, act_type='prelu')
+            model_path = 'weights/realesr-general-x4v3.pth'
+            self.upsampler = RealESRGANer(
+                scale=4, model_path=model_path, model=model, tile=tile, tile_pad=10, pre_pad=0, half=half)
+        elif version == 'Anime - anime6B':
+            model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=6, num_grow_ch=32, scale=4)
+            model_path = 'weights/RealESRGAN_x4plus_anime_6B.pth'
+            self.upsampler = RealESRGANer(
+                scale=4, model_path=model_path, model=model, tile=tile, tile_pad=10, pre_pad=0, half=half)
+        elif version == 'AnimeVideo - v3':
+            model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=16, upscale=4, act_type='prelu')
+            model_path = 'weights/realesr-animevideov3.pth'
+            self.upsampler = RealESRGANer(
+                scale=4, model_path=model_path, model=model, tile=tile, tile_pad=10, pre_pad=0, half=half)
+
+        self.face_enhancer = GFPGANer(
+            model_path='weights/GFPGANv1.4.pth',
+            upscale=scale,
+            arch='clean',
+            channel_multiplier=2,
+            bg_upsampler=self.upsampler)
+
+    def predict(
+        self,
+        img: Path = Input(description='Input'),
+        version: str = Input(
+            description='RealESRGAN version. Please see [Readme] below for more descriptions',
+            choices=['General - RealESRGANplus', 'General - v3', 'Anime - anime6B', 'AnimeVideo - v3'],
+            default='General - v3'),
+        scale: float = Input(description='Rescaling factor', default=2),
+        face_enhance: bool = Input(
+            description='Enhance faces with GFPGAN. Note that it does not work for anime images/vidoes', default=False),
+        tile: int = Input(
+            description=
+            'Tile size. Default is 0, that is no tile. When encountering the out-of-GPU-memory issue, please specify it, e.g., 400 or 200',
+            default=0)
+    ) -> Path:
+        if tile <= 100 or tile is None:
+            tile = 0
+        print(f'img: {img}. version: {version}. scale: {scale}. face_enhance: {face_enhance}. tile: {tile}.')
+        try:
+            extension = os.path.splitext(os.path.basename(str(img)))[1]
+            img = cv2.imread(str(img), cv2.IMREAD_UNCHANGED)
+            if len(img.shape) == 3 and img.shape[2] == 4:
+                img_mode = 'RGBA'
+            elif len(img.shape) == 2:
+                img_mode = None
+                img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
+            else:
+                img_mode = None
+
+            h, w = img.shape[0:2]
+            if h < 300:
+                img = cv2.resize(img, (w * 2, h * 2), interpolation=cv2.INTER_LANCZOS4)
+
+            self.choose_model(scale, version, tile)
+
+            try:
+                if face_enhance:
+                    _, _, output = self.face_enhancer.enhance(
+                        img, has_aligned=False, only_center_face=False, paste_back=True)
+                else:
+                    output, _ = self.upsampler.enhance(img, outscale=scale)
+            except RuntimeError as error:
+                print('Error', error)
+                print('If you encounter CUDA out of memory, try to set "tile" to a smaller size, e.g., 400.')
+
+            if img_mode == 'RGBA':  # RGBA images should be saved in png format
+                extension = 'png'
+            # save_path = f'output/out.{extension}'
+            # cv2.imwrite(save_path, output)
+            out_path = Path(tempfile.mkdtemp()) / f'out.{extension}'
+            cv2.imwrite(str(out_path), output)
+        except Exception as error:
+            print('global exception: ', error)
+        finally:
+            clean_folder('output')
+        return out_path
+
+
+def clean_folder(folder):
+    for filename in os.listdir(folder):
+        file_path = os.path.join(folder, filename)
+        try:
+            if os.path.isfile(file_path) or os.path.islink(file_path):
+                os.unlink(file_path)
+            elif os.path.isdir(file_path):
+                shutil.rmtree(file_path)
+        except Exception as e:
+            print(f'Failed to delete {file_path}. Reason: {e}')

Real-ESRGAN/experiments/pretrained_models/README.md

@@ -0,0 +1 @@
+# Put downloaded pre-trained models here

Real-ESRGAN/gfpgan/weights/detection_Resnet50_Final.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d1de9c2944f2ccddca5f5e010ea5ae64a39845a86311af6fdf30841b0a5a16d
+size 109497761

Real-ESRGAN/gfpgan/weights/parsing_parsenet.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d558d8d0e42c20224f13cf5a29c79eba2d59913419f945545d8cf7b72920de2
+size 85331193

Real-ESRGAN/inference_realesrgan.py

@@ -0,0 +1,166 @@
+import argparse
+import cv2
+import glob
+import os
+from basicsr.archs.rrdbnet_arch import RRDBNet
+from basicsr.utils.download_util import load_file_from_url
+
+from realesrgan import RealESRGANer
+from realesrgan.archs.srvgg_arch import SRVGGNetCompact
+
+
+def main():
+    """Inference demo for Real-ESRGAN.
+    """
+    parser = argparse.ArgumentParser()
+    parser.add_argument('-i', '--input', type=str, default='inputs', help='Input image or folder')
+    parser.add_argument(
+        '-n',
+        '--model_name',
+        type=str,
+        default='RealESRGAN_x4plus',
+        help=('Model names: RealESRGAN_x4plus | RealESRNet_x4plus | RealESRGAN_x4plus_anime_6B | RealESRGAN_x2plus | '
+              'realesr-animevideov3 | realesr-general-x4v3'))
+    parser.add_argument('-o', '--output', type=str, default='results', help='Output folder')
+    parser.add_argument(
+        '-dn',
+        '--denoise_strength',
+        type=float,
+        default=0.5,
+        help=('Denoise strength. 0 for weak denoise (keep noise), 1 for strong denoise ability. '
+              'Only used for the realesr-general-x4v3 model'))
+    parser.add_argument('-s', '--outscale', type=float, default=4, help='The final upsampling scale of the image')
+    parser.add_argument(
+        '--model_path', type=str, default=None, help='[Option] Model path. Usually, you do not need to specify it')
+    parser.add_argument('--suffix', type=str, default='out', help='Suffix of the restored image')
+    parser.add_argument('-t', '--tile', type=int, default=0, help='Tile size, 0 for no tile during testing')
+    parser.add_argument('--tile_pad', type=int, default=10, help='Tile padding')
+    parser.add_argument('--pre_pad', type=int, default=0, help='Pre padding size at each border')
+    parser.add_argument('--face_enhance', action='store_true', help='Use GFPGAN to enhance face')
+    parser.add_argument(
+        '--fp32', action='store_true', help='Use fp32 precision during inference. Default: fp16 (half precision).')
+    parser.add_argument(
+        '--alpha_upsampler',
+        type=str,
+        default='realesrgan',
+        help='The upsampler for the alpha channels. Options: realesrgan | bicubic')
+    parser.add_argument(
+        '--ext',
+        type=str,
+        default='auto',
+        help='Image extension. Options: auto | jpg | png, auto means using the same extension as inputs')
+    parser.add_argument(
+        '-g', '--gpu-id', type=int, default=None, help='gpu device to use (default=None) can be 0,1,2 for multi-gpu')
+
+    args = parser.parse_args()
+
+    # determine models according to model names
+    args.model_name = args.model_name.split('.')[0]
+    if args.model_name == 'RealESRGAN_x4plus':  # x4 RRDBNet model
+        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=4)
+        netscale = 4
+        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth']
+    elif args.model_name == 'RealESRNet_x4plus':  # x4 RRDBNet model
+        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=4)
+        netscale = 4
+        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.1/RealESRNet_x4plus.pth']
+    elif args.model_name == 'RealESRGAN_x4plus_anime_6B':  # x4 RRDBNet model with 6 blocks
+        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=6, num_grow_ch=32, scale=4)
+        netscale = 4
+        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth']
+    elif args.model_name == 'RealESRGAN_x2plus':  # x2 RRDBNet model
+        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=2)
+        netscale = 2
+        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.1/RealESRGAN_x2plus.pth']
+    elif args.model_name == 'realesr-animevideov3':  # x4 VGG-style model (XS size)
+        model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=16, upscale=4, act_type='prelu')
+        netscale = 4
+        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-animevideov3.pth']
+    elif args.model_name == 'realesr-general-x4v3':  # x4 VGG-style model (S size)
+        model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=32, upscale=4, act_type='prelu')
+        netscale = 4
+        file_url = [
+            'https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-wdn-x4v3.pth',
+            'https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth'
+        ]
+
+    # determine model paths
+    if args.model_path is not None:
+        model_path = args.model_path
+    else:
+        model_path = os.path.join('weights', args.model_name + '.pth')
+        if not os.path.isfile(model_path):
+            ROOT_DIR = os.path.dirname(os.path.abspath(__file__))
+            for url in file_url:
+                # model_path will be updated
+                model_path = load_file_from_url(
+                    url=url, model_dir=os.path.join(ROOT_DIR, 'weights'), progress=True, file_name=None)
+
+    # use dni to control the denoise strength
+    dni_weight = None
+    if args.model_name == 'realesr-general-x4v3' and args.denoise_strength != 1:
+        wdn_model_path = model_path.replace('realesr-general-x4v3', 'realesr-general-wdn-x4v3')
+        model_path = [model_path, wdn_model_path]
+        dni_weight = [args.denoise_strength, 1 - args.denoise_strength]
+
+    # restorer
+    upsampler = RealESRGANer(
+        scale=netscale,
+        model_path=model_path,
+        dni_weight=dni_weight,
+        model=model,
+        tile=args.tile,
+        tile_pad=args.tile_pad,
+        pre_pad=args.pre_pad,
+        half=not args.fp32,
+        gpu_id=args.gpu_id)
+
+    if args.face_enhance:  # Use GFPGAN for face enhancement
+        from gfpgan import GFPGANer
+        face_enhancer = GFPGANer(
+            model_path='https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth',
+            upscale=args.outscale,
+            arch='clean',
+            channel_multiplier=2,
+            bg_upsampler=upsampler)
+    os.makedirs(args.output, exist_ok=True)
+
+    if os.path.isfile(args.input):
+        paths = [args.input]
+    else:
+        paths = sorted(glob.glob(os.path.join(args.input, '*')))
+
+    for idx, path in enumerate(paths):
+        imgname, extension = os.path.splitext(os.path.basename(path))
+        print('Testing', idx, imgname)
+
+        img = cv2.imread(path, cv2.IMREAD_UNCHANGED)
+        if len(img.shape) == 3 and img.shape[2] == 4:
+            img_mode = 'RGBA'
+        else:
+            img_mode = None
+
+        try:
+            if args.face_enhance:
+                _, _, output = face_enhancer.enhance(img, has_aligned=False, only_center_face=False, paste_back=True)
+            else:
+                output, _ = upsampler.enhance(img, outscale=args.outscale)
+        except RuntimeError as error:
+            print('Error', error)
+            print('If you encounter CUDA out of memory, try to set --tile with a smaller number.')
+        else:
+            if args.ext == 'auto':
+                extension = extension[1:]
+            else:
+                extension = args.ext
+            if img_mode == 'RGBA':  # RGBA images should be saved in png format
+                extension = 'png'
+            if args.suffix == '':
+                save_path = os.path.join(args.output, f'{imgname}.{extension}')
+            else:
+                save_path = os.path.join(args.output, f'{imgname}_{args.suffix}.{extension}')
+            cv2.imwrite(save_path, output)
+
+
+if __name__ == '__main__':
+    main()

Real-ESRGAN/inference_realesrgan_video.py

@@ -0,0 +1,398 @@
+import argparse
+import cv2
+import glob
+import mimetypes
+import numpy as np
+import os
+import shutil
+import subprocess
+import torch
+from basicsr.archs.rrdbnet_arch import RRDBNet
+from basicsr.utils.download_util import load_file_from_url
+from os import path as osp
+from tqdm import tqdm
+
+from realesrgan import RealESRGANer
+from realesrgan.archs.srvgg_arch import SRVGGNetCompact
+
+try:
+    import ffmpeg
+except ImportError:
+    import pip
+    pip.main(['install', '--user', 'ffmpeg-python'])
+    import ffmpeg
+
+
+def get_video_meta_info(video_path):
+    ret = {}
+    probe = ffmpeg.probe(video_path)
+    video_streams = [stream for stream in probe['streams'] if stream['codec_type'] == 'video']
+    has_audio = any(stream['codec_type'] == 'audio' for stream in probe['streams'])
+    ret['width'] = video_streams[0]['width']
+    ret['height'] = video_streams[0]['height']
+    ret['fps'] = eval(video_streams[0]['avg_frame_rate'])
+    ret['audio'] = ffmpeg.input(video_path).audio if has_audio else None
+    ret['nb_frames'] = int(video_streams[0]['nb_frames'])
+    return ret
+
+
+def get_sub_video(args, num_process, process_idx):
+    if num_process == 1:
+        return args.input
+    meta = get_video_meta_info(args.input)
+    duration = int(meta['nb_frames'] / meta['fps'])
+    part_time = duration // num_process
+    print(f'duration: {duration}, part_time: {part_time}')
+    os.makedirs(osp.join(args.output, f'{args.video_name}_inp_tmp_videos'), exist_ok=True)
+    out_path = osp.join(args.output, f'{args.video_name}_inp_tmp_videos', f'{process_idx:03d}.mp4')
+    cmd = [
+        args.ffmpeg_bin, f'-i {args.input}', '-ss', f'{part_time * process_idx}',
+        f'-to {part_time * (process_idx + 1)}' if process_idx != num_process - 1 else '', '-async 1', out_path, '-y'
+    ]
+    print(' '.join(cmd))
+    subprocess.call(' '.join(cmd), shell=True)
+    return out_path
+
+
+class Reader:
+
+    def __init__(self, args, total_workers=1, worker_idx=0):
+        self.args = args
+        input_type = mimetypes.guess_type(args.input)[0]
+        self.input_type = 'folder' if input_type is None else input_type
+        self.paths = []  # for image&folder type
+        self.audio = None
+        self.input_fps = None
+        if self.input_type.startswith('video'):
+            video_path = get_sub_video(args, total_workers, worker_idx)
+            self.stream_reader = (
+                ffmpeg.input(video_path).output('pipe:', format='rawvideo', pix_fmt='bgr24',
+                                                loglevel='error').run_async(
+                                                    pipe_stdin=True, pipe_stdout=True, cmd=args.ffmpeg_bin))
+            meta = get_video_meta_info(video_path)
+            self.width = meta['width']
+            self.height = meta['height']
+            self.input_fps = meta['fps']
+            self.audio = meta['audio']
+            self.nb_frames = meta['nb_frames']
+
+        else:
+            if self.input_type.startswith('image'):
+                self.paths = [args.input]
+            else:
+                paths = sorted(glob.glob(os.path.join(args.input, '*')))
+                tot_frames = len(paths)
+                num_frame_per_worker = tot_frames // total_workers + (1 if tot_frames % total_workers else 0)
+                self.paths = paths[num_frame_per_worker * worker_idx:num_frame_per_worker * (worker_idx + 1)]
+
+            self.nb_frames = len(self.paths)
+            assert self.nb_frames > 0, 'empty folder'
+            from PIL import Image
+            tmp_img = Image.open(self.paths[0])
+            self.width, self.height = tmp_img.size
+        self.idx = 0
+
+    def get_resolution(self):
+        return self.height, self.width
+
+    def get_fps(self):
+        if self.args.fps is not None:
+            return self.args.fps
+        elif self.input_fps is not None:
+            return self.input_fps
+        return 24
+
+    def get_audio(self):
+        return self.audio
+
+    def __len__(self):
+        return self.nb_frames
+
+    def get_frame_from_stream(self):
+        img_bytes = self.stream_reader.stdout.read(self.width * self.height * 3)  # 3 bytes for one pixel
+        if not img_bytes:
+            return None
+        img = np.frombuffer(img_bytes, np.uint8).reshape([self.height, self.width, 3])
+        return img
+
+    def get_frame_from_list(self):
+        if self.idx >= self.nb_frames:
+            return None
+        img = cv2.imread(self.paths[self.idx])
+        self.idx += 1
+        return img
+
+    def get_frame(self):
+        if self.input_type.startswith('video'):
+            return self.get_frame_from_stream()
+        else:
+            return self.get_frame_from_list()
+
+    def close(self):
+        if self.input_type.startswith('video'):
+            self.stream_reader.stdin.close()
+            self.stream_reader.wait()
+
+
+class Writer:
+
+    def __init__(self, args, audio, height, width, video_save_path, fps):
+        out_width, out_height = int(width * args.outscale), int(height * args.outscale)
+        if out_height > 2160:
+            print('You are generating video that is larger than 4K, which will be very slow due to IO speed.',
+                  'We highly recommend to decrease the outscale(aka, -s).')
+
+        if audio is not None:
+            self.stream_writer = (
+                ffmpeg.input('pipe:', format='rawvideo', pix_fmt='bgr24', s=f'{out_width}x{out_height}',
+                             framerate=fps).output(
+                                 audio,
+                                 video_save_path,
+                                 pix_fmt='yuv420p',
+                                 vcodec='libx264',
+                                 loglevel='error',
+                                 acodec='copy').overwrite_output().run_async(
+                                     pipe_stdin=True, pipe_stdout=True, cmd=args.ffmpeg_bin))
+        else:
+            self.stream_writer = (
+                ffmpeg.input('pipe:', format='rawvideo', pix_fmt='bgr24', s=f'{out_width}x{out_height}',
+                             framerate=fps).output(
+                                 video_save_path, pix_fmt='yuv420p', vcodec='libx264',
+                                 loglevel='error').overwrite_output().run_async(
+                                     pipe_stdin=True, pipe_stdout=True, cmd=args.ffmpeg_bin))
+
+    def write_frame(self, frame):
+        frame = frame.astype(np.uint8).tobytes()
+        self.stream_writer.stdin.write(frame)
+
+    def close(self):
+        self.stream_writer.stdin.close()
+        self.stream_writer.wait()
+
+
+def inference_video(args, video_save_path, device=None, total_workers=1, worker_idx=0):
+    # ---------------------- determine models according to model names ---------------------- #
+    args.model_name = args.model_name.split('.pth')[0]
+    if args.model_name == 'RealESRGAN_x4plus':  # x4 RRDBNet model
+        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=4)
+        netscale = 4
+        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth']
+    elif args.model_name == 'RealESRNet_x4plus':  # x4 RRDBNet model
+        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=4)
+        netscale = 4
+        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.1/RealESRNet_x4plus.pth']
+    elif args.model_name == 'RealESRGAN_x4plus_anime_6B':  # x4 RRDBNet model with 6 blocks
+        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=6, num_grow_ch=32, scale=4)
+        netscale = 4
+        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.2.4/RealESRGAN_x4plus_anime_6B.pth']
+    elif args.model_name == 'RealESRGAN_x2plus':  # x2 RRDBNet model
+        model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64, num_block=23, num_grow_ch=32, scale=2)
+        netscale = 2
+        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.1/RealESRGAN_x2plus.pth']
+    elif args.model_name == 'realesr-animevideov3':  # x4 VGG-style model (XS size)
+        model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=16, upscale=4, act_type='prelu')
+        netscale = 4
+        file_url = ['https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-animevideov3.pth']
+    elif args.model_name == 'realesr-general-x4v3':  # x4 VGG-style model (S size)
+        model = SRVGGNetCompact(num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=32, upscale=4, act_type='prelu')
+        netscale = 4
+        file_url = [
+            'https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-wdn-x4v3.pth',
+            'https://github.com/xinntao/Real-ESRGAN/releases/download/v0.2.5.0/realesr-general-x4v3.pth'
+        ]
+
+    # ---------------------- determine model paths ---------------------- #
+    model_path = os.path.join('weights', args.model_name + '.pth')
+    if not os.path.isfile(model_path):
+        ROOT_DIR = os.path.dirname(os.path.abspath(__file__))
+        for url in file_url:
+            # model_path will be updated
+            model_path = load_file_from_url(
+                url=url, model_dir=os.path.join(ROOT_DIR, 'weights'), progress=True, file_name=None)
+
+    # use dni to control the denoise strength
+    dni_weight = None
+    if args.model_name == 'realesr-general-x4v3' and args.denoise_strength != 1:
+        wdn_model_path = model_path.replace('realesr-general-x4v3', 'realesr-general-wdn-x4v3')
+        model_path = [model_path, wdn_model_path]
+        dni_weight = [args.denoise_strength, 1 - args.denoise_strength]
+
+    # restorer
+    upsampler = RealESRGANer(
+        scale=netscale,
+        model_path=model_path,
+        dni_weight=dni_weight,
+        model=model,
+        tile=args.tile,
+        tile_pad=args.tile_pad,
+        pre_pad=args.pre_pad,
+        half=not args.fp32,
+        device=device,
+    )
+
+    if 'anime' in args.model_name and args.face_enhance:
+        print('face_enhance is not supported in anime models, we turned this option off for you. '
+              'if you insist on turning it on, please manually comment the relevant lines of code.')
+        args.face_enhance = False
+
+    if args.face_enhance:  # Use GFPGAN for face enhancement
+        from gfpgan import GFPGANer
+        face_enhancer = GFPGANer(
+            model_path='https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth',
+            upscale=args.outscale,
+            arch='clean',
+            channel_multiplier=2,
+            bg_upsampler=upsampler)  # TODO support custom device
+    else:
+        face_enhancer = None
+
+    reader = Reader(args, total_workers, worker_idx)
+    audio = reader.get_audio()
+    height, width = reader.get_resolution()
+    fps = reader.get_fps()
+    writer = Writer(args, audio, height, width, video_save_path, fps)
+
+    pbar = tqdm(total=len(reader), unit='frame', desc='inference')
+    while True:
+        img = reader.get_frame()
+        if img is None:
+            break
+
+        try:
+            if args.face_enhance:
+                _, _, output = face_enhancer.enhance(img, has_aligned=False, only_center_face=False, paste_back=True)
+            else:
+                output, _ = upsampler.enhance(img, outscale=args.outscale)
+        except RuntimeError as error:
+            print('Error', error)
+            print('If you encounter CUDA out of memory, try to set --tile with a smaller number.')
+        else:
+            writer.write_frame(output)
+
+        torch.cuda.synchronize(device)
+        pbar.update(1)
+
+    reader.close()
+    writer.close()
+
+
+def run(args):
+    args.video_name = osp.splitext(os.path.basename(args.input))[0]
+    video_save_path = osp.join(args.output, f'{args.video_name}_{args.suffix}.mp4')
+
+    if args.extract_frame_first:
+        tmp_frames_folder = osp.join(args.output, f'{args.video_name}_inp_tmp_frames')
+        os.makedirs(tmp_frames_folder, exist_ok=True)
+        os.system(f'ffmpeg -i {args.input} -qscale:v 1 -qmin 1 -qmax 1 -vsync 0  {tmp_frames_folder}/frame%08d.png')
+        args.input = tmp_frames_folder
+
+    num_gpus = torch.cuda.device_count()
+    num_process = num_gpus * args.num_process_per_gpu
+    if num_process == 1:
+        inference_video(args, video_save_path)
+        return
+
+    ctx = torch.multiprocessing.get_context('spawn')
+    pool = ctx.Pool(num_process)
+    os.makedirs(osp.join(args.output, f'{args.video_name}_out_tmp_videos'), exist_ok=True)
+    pbar = tqdm(total=num_process, unit='sub_video', desc='inference')
+    for i in range(num_process):
+        sub_video_save_path = osp.join(args.output, f'{args.video_name}_out_tmp_videos', f'{i:03d}.mp4')
+        pool.apply_async(
+            inference_video,
+            args=(args, sub_video_save_path, torch.device(i % num_gpus), num_process, i),
+            callback=lambda arg: pbar.update(1))
+    pool.close()
+    pool.join()
+
+    # combine sub videos
+    # prepare vidlist.txt
+    with open(f'{args.output}/{args.video_name}_vidlist.txt', 'w') as f:
+        for i in range(num_process):
+            f.write(f'file \'{args.video_name}_out_tmp_videos/{i:03d}.mp4\'\n')
+
+    cmd = [
+        args.ffmpeg_bin, '-f', 'concat', '-safe', '0', '-i', f'{args.output}/{args.video_name}_vidlist.txt', '-c',
+        'copy', f'{video_save_path}'
+    ]
+    print(' '.join(cmd))
+    subprocess.call(cmd)
+    shutil.rmtree(osp.join(args.output, f'{args.video_name}_out_tmp_videos'))
+    if osp.exists(osp.join(args.output, f'{args.video_name}_inp_tmp_videos')):
+        shutil.rmtree(osp.join(args.output, f'{args.video_name}_inp_tmp_videos'))
+    os.remove(f'{args.output}/{args.video_name}_vidlist.txt')
+
+
+def main():
+    """Inference demo for Real-ESRGAN.
+    It mainly for restoring anime videos.
+
+    """
+    parser = argparse.ArgumentParser()
+    parser.add_argument('-i', '--input', type=str, default='inputs', help='Input video, image or folder')
+    parser.add_argument(
+        '-n',
+        '--model_name',
+        type=str,
+        default='realesr-animevideov3',
+        help=('Model names: realesr-animevideov3 | RealESRGAN_x4plus_anime_6B | RealESRGAN_x4plus | RealESRNet_x4plus |'
+              ' RealESRGAN_x2plus | realesr-general-x4v3'
+              'Default:realesr-animevideov3'))
+    parser.add_argument('-o', '--output', type=str, default='results', help='Output folder')
+    parser.add_argument(
+        '-dn',
+        '--denoise_strength',
+        type=float,
+        default=0.5,
+        help=('Denoise strength. 0 for weak denoise (keep noise), 1 for strong denoise ability. '
+              'Only used for the realesr-general-x4v3 model'))
+    parser.add_argument('-s', '--outscale', type=float, default=4, help='The final upsampling scale of the image')
+    parser.add_argument('--suffix', type=str, default='out', help='Suffix of the restored video')
+    parser.add_argument('-t', '--tile', type=int, default=0, help='Tile size, 0 for no tile during testing')
+    parser.add_argument('--tile_pad', type=int, default=10, help='Tile padding')
+    parser.add_argument('--pre_pad', type=int, default=0, help='Pre padding size at each border')
+    parser.add_argument('--face_enhance', action='store_true', help='Use GFPGAN to enhance face')
+    parser.add_argument(
+        '--fp32', action='store_true', help='Use fp32 precision during inference. Default: fp16 (half precision).')
+    parser.add_argument('--fps', type=float, default=None, help='FPS of the output video')
+    parser.add_argument('--ffmpeg_bin', type=str, default='ffmpeg', help='The path to ffmpeg')
+    parser.add_argument('--extract_frame_first', action='store_true')
+    parser.add_argument('--num_process_per_gpu', type=int, default=1)
+
+    parser.add_argument(
+        '--alpha_upsampler',
+        type=str,
+        default='realesrgan',
+        help='The upsampler for the alpha channels. Options: realesrgan | bicubic')
+    parser.add_argument(
+        '--ext',
+        type=str,
+        default='auto',
+        help='Image extension. Options: auto | jpg | png, auto means using the same extension as inputs')
+    args = parser.parse_args()
+
+    args.input = args.input.rstrip('/').rstrip('\\')
+    os.makedirs(args.output, exist_ok=True)
+
+    if mimetypes.guess_type(args.input)[0] is not None and mimetypes.guess_type(args.input)[0].startswith('video'):
+        is_video = True
+    else:
+        is_video = False
+
+    if is_video and args.input.endswith('.flv'):
+        mp4_path = args.input.replace('.flv', '.mp4')
+        os.system(f'ffmpeg -i {args.input} -codec copy {mp4_path}')
+        args.input = mp4_path
+
+    if args.extract_frame_first and not is_video:
+        args.extract_frame_first = False
+
+    run(args)
+
+    if args.extract_frame_first:
+        tmp_frames_folder = osp.join(args.output, f'{args.video_name}_inp_tmp_frames')
+        shutil.rmtree(tmp_frames_folder)
+
+
+if __name__ == '__main__':
+    main()

Real-ESRGAN/options/finetune_realesrgan_x4plus.yml

@@ -0,0 +1,188 @@
+# general settings
+name: finetune_RealESRGANx4plus_400k
+model_type: RealESRGANModel
+scale: 4
+num_gpu: auto
+manual_seed: 0
+
+# ----------------- options for synthesizing training data in RealESRGANModel ----------------- #
+# USM the ground-truth
+l1_gt_usm: True
+percep_gt_usm: True
+gan_gt_usm: False
+
+# the first degradation process
+resize_prob: [0.2, 0.7, 0.1]  # up, down, keep
+resize_range: [0.15, 1.5]
+gaussian_noise_prob: 0.5
+noise_range: [1, 30]
+poisson_scale_range: [0.05, 3]
+gray_noise_prob: 0.4
+jpeg_range: [30, 95]
+
+# the second degradation process
+second_blur_prob: 0.8
+resize_prob2: [0.3, 0.4, 0.3]  # up, down, keep
+resize_range2: [0.3, 1.2]
+gaussian_noise_prob2: 0.5
+noise_range2: [1, 25]
+poisson_scale_range2: [0.05, 2.5]
+gray_noise_prob2: 0.4
+jpeg_range2: [30, 95]
+
+gt_size: 256
+queue_size: 180
+
+# dataset and data loader settings
+datasets:
+  train:
+    name: DF2K+OST
+    type: RealESRGANDataset
+    dataroot_gt: datasets/DF2K
+    meta_info: datasets/DF2K/meta_info/meta_info_DF2Kmultiscale+OST_sub.txt
+    io_backend:
+      type: disk
+
+    blur_kernel_size: 21
+    kernel_list: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
+    kernel_prob: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
+    sinc_prob: 0.1
+    blur_sigma: [0.2, 3]
+    betag_range: [0.5, 4]
+    betap_range: [1, 2]
+
+    blur_kernel_size2: 21
+    kernel_list2: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
+    kernel_prob2: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
+    sinc_prob2: 0.1
+    blur_sigma2: [0.2, 1.5]
+    betag_range2: [0.5, 4]
+    betap_range2: [1, 2]
+
+    final_sinc_prob: 0.8
+
+    gt_size: 256
+    use_hflip: True
+    use_rot: False
+
+    # data loader
+    use_shuffle: true
+    num_worker_per_gpu: 5
+    batch_size_per_gpu: 12
+    dataset_enlarge_ratio: 1
+    prefetch_mode: ~
+
+  # Uncomment these for validation
+  # val:
+  #   name: validation
+  #   type: PairedImageDataset
+  #   dataroot_gt: path_to_gt
+  #   dataroot_lq: path_to_lq
+  #   io_backend:
+  #     type: disk
+
+# network structures
+network_g:
+  type: RRDBNet
+  num_in_ch: 3
+  num_out_ch: 3
+  num_feat: 64
+  num_block: 23
+  num_grow_ch: 32
+
+network_d:
+  type: UNetDiscriminatorSN
+  num_in_ch: 3
+  num_feat: 64
+  skip_connection: True
+
+# path
+path:
+  # use the pre-trained Real-ESRNet model
+  pretrain_network_g: experiments/pretrained_models/RealESRNet_x4plus.pth
+  param_key_g: params_ema
+  strict_load_g: true
+  pretrain_network_d: experiments/pretrained_models/RealESRGAN_x4plus_netD.pth
+  param_key_d: params
+  strict_load_d: true
+  resume_state: ~
+
+# training settings
+train:
+  ema_decay: 0.999
+  optim_g:
+    type: Adam
+    lr: !!float 1e-4
+    weight_decay: 0
+    betas: [0.9, 0.99]
+  optim_d:
+    type: Adam
+    lr: !!float 1e-4
+    weight_decay: 0
+    betas: [0.9, 0.99]
+
+  scheduler:
+    type: MultiStepLR
+    milestones: [400000]
+    gamma: 0.5
+
+  total_iter: 400000
+  warmup_iter: -1  # no warm up
+
+  # losses
+  pixel_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+  # perceptual loss (content and style losses)
+  perceptual_opt:
+    type: PerceptualLoss
+    layer_weights:
+      # before relu
+      'conv1_2': 0.1
+      'conv2_2': 0.1
+      'conv3_4': 1
+      'conv4_4': 1
+      'conv5_4': 1
+    vgg_type: vgg19
+    use_input_norm: true
+    perceptual_weight: !!float 1.0
+    style_weight: 0
+    range_norm: false
+    criterion: l1
+  # gan loss
+  gan_opt:
+    type: GANLoss
+    gan_type: vanilla
+    real_label_val: 1.0
+    fake_label_val: 0.0
+    loss_weight: !!float 1e-1
+
+  net_d_iters: 1
+  net_d_init_iters: 0
+
+# Uncomment these for validation
+# validation settings
+# val:
+#   val_freq: !!float 5e3
+#   save_img: True
+
+#   metrics:
+#     psnr: # metric name
+#       type: calculate_psnr
+#       crop_border: 4
+#       test_y_channel: false
+
+# logging settings
+logger:
+  print_freq: 100
+  save_checkpoint_freq: !!float 5e3
+  use_tb_logger: true
+  wandb:
+    project: ~
+    resume_id: ~
+
+# dist training settings
+dist_params:
+  backend: nccl
+  port: 29500

Real-ESRGAN/options/finetune_realesrgan_x4plus_pairdata.yml

@@ -0,0 +1,150 @@
+# general settings
+name: finetune_RealESRGANx4plus_400k_pairdata
+model_type: RealESRGANModel
+scale: 4
+num_gpu: auto
+manual_seed: 0
+
+# USM the ground-truth
+l1_gt_usm: True
+percep_gt_usm: True
+gan_gt_usm: False
+
+high_order_degradation: False # do not use the high-order degradation generation process
+
+# dataset and data loader settings
+datasets:
+  train:
+    name: DIV2K
+    type: RealESRGANPairedDataset
+    dataroot_gt: datasets/DF2K
+    dataroot_lq: datasets/DF2K
+    meta_info: datasets/DF2K/meta_info/meta_info_DIV2K_sub_pair.txt
+    io_backend:
+      type: disk
+
+    gt_size: 256
+    use_hflip: True
+    use_rot: False
+
+    # data loader
+    use_shuffle: true
+    num_worker_per_gpu: 5
+    batch_size_per_gpu: 12
+    dataset_enlarge_ratio: 1
+    prefetch_mode: ~
+
+  # Uncomment these for validation
+  # val:
+  #   name: validation
+  #   type: PairedImageDataset
+  #   dataroot_gt: path_to_gt
+  #   dataroot_lq: path_to_lq
+  #   io_backend:
+  #     type: disk
+
+# network structures
+network_g:
+  type: RRDBNet
+  num_in_ch: 3
+  num_out_ch: 3
+  num_feat: 64
+  num_block: 23
+  num_grow_ch: 32
+
+network_d:
+  type: UNetDiscriminatorSN
+  num_in_ch: 3
+  num_feat: 64
+  skip_connection: True
+
+# path
+path:
+  # use the pre-trained Real-ESRNet model
+  pretrain_network_g: experiments/pretrained_models/RealESRNet_x4plus.pth
+  param_key_g: params_ema
+  strict_load_g: true
+  pretrain_network_d: experiments/pretrained_models/RealESRGAN_x4plus_netD.pth
+  param_key_d: params
+  strict_load_d: true
+  resume_state: ~
+
+# training settings
+train:
+  ema_decay: 0.999
+  optim_g:
+    type: Adam
+    lr: !!float 1e-4
+    weight_decay: 0
+    betas: [0.9, 0.99]
+  optim_d:
+    type: Adam
+    lr: !!float 1e-4
+    weight_decay: 0
+    betas: [0.9, 0.99]
+
+  scheduler:
+    type: MultiStepLR
+    milestones: [400000]
+    gamma: 0.5
+
+  total_iter: 400000
+  warmup_iter: -1  # no warm up
+
+  # losses
+  pixel_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+  # perceptual loss (content and style losses)
+  perceptual_opt:
+    type: PerceptualLoss
+    layer_weights:
+      # before relu
+      'conv1_2': 0.1
+      'conv2_2': 0.1
+      'conv3_4': 1
+      'conv4_4': 1
+      'conv5_4': 1
+    vgg_type: vgg19
+    use_input_norm: true
+    perceptual_weight: !!float 1.0
+    style_weight: 0
+    range_norm: false
+    criterion: l1
+  # gan loss
+  gan_opt:
+    type: GANLoss
+    gan_type: vanilla
+    real_label_val: 1.0
+    fake_label_val: 0.0
+    loss_weight: !!float 1e-1
+
+  net_d_iters: 1
+  net_d_init_iters: 0
+
+# Uncomment these for validation
+# validation settings
+# val:
+#   val_freq: !!float 5e3
+#   save_img: True
+
+#   metrics:
+#     psnr: # metric name
+#       type: calculate_psnr
+#       crop_border: 4
+#       test_y_channel: false
+
+# logging settings
+logger:
+  print_freq: 100
+  save_checkpoint_freq: !!float 5e3
+  use_tb_logger: true
+  wandb:
+    project: ~
+    resume_id: ~
+
+# dist training settings
+dist_params:
+  backend: nccl
+  port: 29500

Real-ESRGAN/options/train_realesrgan_x2plus.yml

@@ -0,0 +1,186 @@
+# general settings
+name: train_RealESRGANx2plus_400k_B12G4
+model_type: RealESRGANModel
+scale: 2
+num_gpu: auto  # auto: can infer from your visible devices automatically. official: 4 GPUs
+manual_seed: 0
+
+# ----------------- options for synthesizing training data in RealESRGANModel ----------------- #
+# USM the ground-truth
+l1_gt_usm: True
+percep_gt_usm: True
+gan_gt_usm: False
+
+# the first degradation process
+resize_prob: [0.2, 0.7, 0.1]  # up, down, keep
+resize_range: [0.15, 1.5]
+gaussian_noise_prob: 0.5
+noise_range: [1, 30]
+poisson_scale_range: [0.05, 3]
+gray_noise_prob: 0.4
+jpeg_range: [30, 95]
+
+# the second degradation process
+second_blur_prob: 0.8
+resize_prob2: [0.3, 0.4, 0.3]  # up, down, keep
+resize_range2: [0.3, 1.2]
+gaussian_noise_prob2: 0.5
+noise_range2: [1, 25]
+poisson_scale_range2: [0.05, 2.5]
+gray_noise_prob2: 0.4
+jpeg_range2: [30, 95]
+
+gt_size: 256
+queue_size: 180
+
+# dataset and data loader settings
+datasets:
+  train:
+    name: DF2K+OST
+    type: RealESRGANDataset
+    dataroot_gt: datasets/DF2K
+    meta_info: datasets/DF2K/meta_info/meta_info_DF2Kmultiscale+OST_sub.txt
+    io_backend:
+      type: disk
+
+    blur_kernel_size: 21
+    kernel_list: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
+    kernel_prob: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
+    sinc_prob: 0.1
+    blur_sigma: [0.2, 3]
+    betag_range: [0.5, 4]
+    betap_range: [1, 2]
+
+    blur_kernel_size2: 21
+    kernel_list2: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
+    kernel_prob2: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
+    sinc_prob2: 0.1
+    blur_sigma2: [0.2, 1.5]
+    betag_range2: [0.5, 4]
+    betap_range2: [1, 2]
+
+    final_sinc_prob: 0.8
+
+    gt_size: 256
+    use_hflip: True
+    use_rot: False
+
+    # data loader
+    use_shuffle: true
+    num_worker_per_gpu: 5
+    batch_size_per_gpu: 12
+    dataset_enlarge_ratio: 1
+    prefetch_mode: ~
+
+  # Uncomment these for validation
+  # val:
+  #   name: validation
+  #   type: PairedImageDataset
+  #   dataroot_gt: path_to_gt
+  #   dataroot_lq: path_to_lq
+  #   io_backend:
+  #     type: disk
+
+# network structures
+network_g:
+  type: RRDBNet
+  num_in_ch: 3
+  num_out_ch: 3
+  num_feat: 64
+  num_block: 23
+  num_grow_ch: 32
+  scale: 2
+
+network_d:
+  type: UNetDiscriminatorSN
+  num_in_ch: 3
+  num_feat: 64
+  skip_connection: True
+
+# path
+path:
+  # use the pre-trained Real-ESRNet model
+  pretrain_network_g: experiments/pretrained_models/RealESRNet_x2plus.pth
+  param_key_g: params_ema
+  strict_load_g: true
+  resume_state: ~
+
+# training settings
+train:
+  ema_decay: 0.999
+  optim_g:
+    type: Adam
+    lr: !!float 1e-4
+    weight_decay: 0
+    betas: [0.9, 0.99]
+  optim_d:
+    type: Adam
+    lr: !!float 1e-4
+    weight_decay: 0
+    betas: [0.9, 0.99]
+
+  scheduler:
+    type: MultiStepLR
+    milestones: [400000]
+    gamma: 0.5
+
+  total_iter: 400000
+  warmup_iter: -1  # no warm up
+
+  # losses
+  pixel_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+  # perceptual loss (content and style losses)
+  perceptual_opt:
+    type: PerceptualLoss
+    layer_weights:
+      # before relu
+      'conv1_2': 0.1
+      'conv2_2': 0.1
+      'conv3_4': 1
+      'conv4_4': 1
+      'conv5_4': 1
+    vgg_type: vgg19
+    use_input_norm: true
+    perceptual_weight: !!float 1.0
+    style_weight: 0
+    range_norm: false
+    criterion: l1
+  # gan loss
+  gan_opt:
+    type: GANLoss
+    gan_type: vanilla
+    real_label_val: 1.0
+    fake_label_val: 0.0
+    loss_weight: !!float 1e-1
+
+  net_d_iters: 1
+  net_d_init_iters: 0
+
+# Uncomment these for validation
+# validation settings
+# val:
+#   val_freq: !!float 5e3
+#   save_img: True
+
+#   metrics:
+#     psnr: # metric name
+#       type: calculate_psnr
+#       crop_border: 4
+#       test_y_channel: false
+
+# logging settings
+logger:
+  print_freq: 100
+  save_checkpoint_freq: !!float 5e3
+  use_tb_logger: true
+  wandb:
+    project: ~
+    resume_id: ~
+
+# dist training settings
+dist_params:
+  backend: nccl
+  port: 29500

Real-ESRGAN/options/train_realesrgan_x4plus.yml

@@ -0,0 +1,185 @@
+# general settings
+name: train_RealESRGANx4plus_400k_B12G4
+model_type: RealESRGANModel
+scale: 4
+num_gpu: auto  # auto: can infer from your visible devices automatically. official: 4 GPUs
+manual_seed: 0
+
+# ----------------- options for synthesizing training data in RealESRGANModel ----------------- #
+# USM the ground-truth
+l1_gt_usm: True
+percep_gt_usm: True
+gan_gt_usm: False
+
+# the first degradation process
+resize_prob: [0.2, 0.7, 0.1]  # up, down, keep
+resize_range: [0.15, 1.5]
+gaussian_noise_prob: 0.5
+noise_range: [1, 30]
+poisson_scale_range: [0.05, 3]
+gray_noise_prob: 0.4
+jpeg_range: [30, 95]
+
+# the second degradation process
+second_blur_prob: 0.8
+resize_prob2: [0.3, 0.4, 0.3]  # up, down, keep
+resize_range2: [0.3, 1.2]
+gaussian_noise_prob2: 0.5
+noise_range2: [1, 25]
+poisson_scale_range2: [0.05, 2.5]
+gray_noise_prob2: 0.4
+jpeg_range2: [30, 95]
+
+gt_size: 256
+queue_size: 180
+
+# dataset and data loader settings
+datasets:
+  train:
+    name: DF2K+OST
+    type: RealESRGANDataset
+    dataroot_gt: datasets/DF2K
+    meta_info: datasets/DF2K/meta_info/meta_info_DF2Kmultiscale+OST_sub.txt
+    io_backend:
+      type: disk
+
+    blur_kernel_size: 21
+    kernel_list: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
+    kernel_prob: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
+    sinc_prob: 0.1
+    blur_sigma: [0.2, 3]
+    betag_range: [0.5, 4]
+    betap_range: [1, 2]
+
+    blur_kernel_size2: 21
+    kernel_list2: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
+    kernel_prob2: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
+    sinc_prob2: 0.1
+    blur_sigma2: [0.2, 1.5]
+    betag_range2: [0.5, 4]
+    betap_range2: [1, 2]
+
+    final_sinc_prob: 0.8
+
+    gt_size: 256
+    use_hflip: True
+    use_rot: False
+
+    # data loader
+    use_shuffle: true
+    num_worker_per_gpu: 5
+    batch_size_per_gpu: 12
+    dataset_enlarge_ratio: 1
+    prefetch_mode: ~
+
+  # Uncomment these for validation
+  # val:
+  #   name: validation
+  #   type: PairedImageDataset
+  #   dataroot_gt: path_to_gt
+  #   dataroot_lq: path_to_lq
+  #   io_backend:
+  #     type: disk
+
+# network structures
+network_g:
+  type: RRDBNet
+  num_in_ch: 3
+  num_out_ch: 3
+  num_feat: 64
+  num_block: 23
+  num_grow_ch: 32
+
+network_d:
+  type: UNetDiscriminatorSN
+  num_in_ch: 3
+  num_feat: 64
+  skip_connection: True
+
+# path
+path:
+  # use the pre-trained Real-ESRNet model
+  pretrain_network_g: experiments/pretrained_models/RealESRNet_x4plus.pth
+  param_key_g: params_ema
+  strict_load_g: true
+  resume_state: ~
+
+# training settings
+train:
+  ema_decay: 0.999
+  optim_g:
+    type: Adam
+    lr: !!float 1e-4
+    weight_decay: 0
+    betas: [0.9, 0.99]
+  optim_d:
+    type: Adam
+    lr: !!float 1e-4
+    weight_decay: 0
+    betas: [0.9, 0.99]
+
+  scheduler:
+    type: MultiStepLR
+    milestones: [400000]
+    gamma: 0.5
+
+  total_iter: 400000
+  warmup_iter: -1  # no warm up
+
+  # losses
+  pixel_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+  # perceptual loss (content and style losses)
+  perceptual_opt:
+    type: PerceptualLoss
+    layer_weights:
+      # before relu
+      'conv1_2': 0.1
+      'conv2_2': 0.1
+      'conv3_4': 1
+      'conv4_4': 1
+      'conv5_4': 1
+    vgg_type: vgg19
+    use_input_norm: true
+    perceptual_weight: !!float 1.0
+    style_weight: 0
+    range_norm: false
+    criterion: l1
+  # gan loss
+  gan_opt:
+    type: GANLoss
+    gan_type: vanilla
+    real_label_val: 1.0
+    fake_label_val: 0.0
+    loss_weight: !!float 1e-1
+
+  net_d_iters: 1
+  net_d_init_iters: 0
+
+# Uncomment these for validation
+# validation settings
+# val:
+#   val_freq: !!float 5e3
+#   save_img: True
+
+#   metrics:
+#     psnr: # metric name
+#       type: calculate_psnr
+#       crop_border: 4
+#       test_y_channel: false
+
+# logging settings
+logger:
+  print_freq: 100
+  save_checkpoint_freq: !!float 5e3
+  use_tb_logger: true
+  wandb:
+    project: ~
+    resume_id: ~
+
+# dist training settings
+dist_params:
+  backend: nccl
+  port: 29500

Real-ESRGAN/options/train_realesrnet_x2plus.yml

@@ -0,0 +1,145 @@
+# general settings
+name: train_RealESRNetx2plus_1000k_B12G4
+model_type: RealESRNetModel
+scale: 2
+num_gpu: auto  # auto: can infer from your visible devices automatically. official: 4 GPUs
+manual_seed: 0
+
+# ----------------- options for synthesizing training data in RealESRNetModel ----------------- #
+gt_usm: True  # USM the ground-truth
+
+# the first degradation process
+resize_prob: [0.2, 0.7, 0.1]  # up, down, keep
+resize_range: [0.15, 1.5]
+gaussian_noise_prob: 0.5
+noise_range: [1, 30]
+poisson_scale_range: [0.05, 3]
+gray_noise_prob: 0.4
+jpeg_range: [30, 95]
+
+# the second degradation process
+second_blur_prob: 0.8
+resize_prob2: [0.3, 0.4, 0.3]  # up, down, keep
+resize_range2: [0.3, 1.2]
+gaussian_noise_prob2: 0.5
+noise_range2: [1, 25]
+poisson_scale_range2: [0.05, 2.5]
+gray_noise_prob2: 0.4
+jpeg_range2: [30, 95]
+
+gt_size: 256
+queue_size: 180
+
+# dataset and data loader settings
+datasets:
+  train:
+    name: DF2K+OST
+    type: RealESRGANDataset
+    dataroot_gt: datasets/DF2K
+    meta_info: datasets/DF2K/meta_info/meta_info_DF2Kmultiscale+OST_sub.txt
+    io_backend:
+      type: disk
+
+    blur_kernel_size: 21
+    kernel_list: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
+    kernel_prob: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
+    sinc_prob: 0.1
+    blur_sigma: [0.2, 3]
+    betag_range: [0.5, 4]
+    betap_range: [1, 2]
+
+    blur_kernel_size2: 21
+    kernel_list2: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
+    kernel_prob2: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
+    sinc_prob2: 0.1
+    blur_sigma2: [0.2, 1.5]
+    betag_range2: [0.5, 4]
+    betap_range2: [1, 2]
+
+    final_sinc_prob: 0.8
+
+    gt_size: 256
+    use_hflip: True
+    use_rot: False
+
+    # data loader
+    use_shuffle: true
+    num_worker_per_gpu: 5
+    batch_size_per_gpu: 12
+    dataset_enlarge_ratio: 1
+    prefetch_mode: ~
+
+  # Uncomment these for validation
+  # val:
+  #   name: validation
+  #   type: PairedImageDataset
+  #   dataroot_gt: path_to_gt
+  #   dataroot_lq: path_to_lq
+  #   io_backend:
+  #     type: disk
+
+# network structures
+network_g:
+  type: RRDBNet
+  num_in_ch: 3
+  num_out_ch: 3
+  num_feat: 64
+  num_block: 23
+  num_grow_ch: 32
+  scale: 2
+
+# path
+path:
+  pretrain_network_g: experiments/pretrained_models/RealESRGAN_x4plus.pth
+  param_key_g: params_ema
+  strict_load_g: False
+  resume_state: ~
+
+# training settings
+train:
+  ema_decay: 0.999
+  optim_g:
+    type: Adam
+    lr: !!float 2e-4
+    weight_decay: 0
+    betas: [0.9, 0.99]
+
+  scheduler:
+    type: MultiStepLR
+    milestones: [1000000]
+    gamma: 0.5
+
+  total_iter: 1000000
+  warmup_iter: -1  # no warm up
+
+  # losses
+  pixel_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+
+# Uncomment these for validation
+# validation settings
+# val:
+#   val_freq: !!float 5e3
+#   save_img: True
+
+#   metrics:
+#     psnr: # metric name
+#       type: calculate_psnr
+#       crop_border: 4
+#       test_y_channel: false
+
+# logging settings
+logger:
+  print_freq: 100
+  save_checkpoint_freq: !!float 5e3
+  use_tb_logger: true
+  wandb:
+    project: ~
+    resume_id: ~
+
+# dist training settings
+dist_params:
+  backend: nccl
+  port: 29500

Real-ESRGAN/options/train_realesrnet_x4plus.yml

@@ -0,0 +1,144 @@
+# general settings
+name: train_RealESRNetx4plus_1000k_B12G4
+model_type: RealESRNetModel
+scale: 4
+num_gpu: auto  # auto: can infer from your visible devices automatically. official: 4 GPUs
+manual_seed: 0
+
+# ----------------- options for synthesizing training data in RealESRNetModel ----------------- #
+gt_usm: True  # USM the ground-truth
+
+# the first degradation process
+resize_prob: [0.2, 0.7, 0.1]  # up, down, keep
+resize_range: [0.15, 1.5]
+gaussian_noise_prob: 0.5
+noise_range: [1, 30]
+poisson_scale_range: [0.05, 3]
+gray_noise_prob: 0.4
+jpeg_range: [30, 95]
+
+# the second degradation process
+second_blur_prob: 0.8
+resize_prob2: [0.3, 0.4, 0.3]  # up, down, keep
+resize_range2: [0.3, 1.2]
+gaussian_noise_prob2: 0.5
+noise_range2: [1, 25]
+poisson_scale_range2: [0.05, 2.5]
+gray_noise_prob2: 0.4
+jpeg_range2: [30, 95]
+
+gt_size: 256
+queue_size: 180
+
+# dataset and data loader settings
+datasets:
+  train:
+    name: DF2K+OST
+    type: RealESRGANDataset
+    dataroot_gt: datasets/DF2K
+    meta_info: datasets/DF2K/meta_info/meta_info_DF2Kmultiscale+OST_sub.txt
+    io_backend:
+      type: disk
+
+    blur_kernel_size: 21
+    kernel_list: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
+    kernel_prob: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
+    sinc_prob: 0.1
+    blur_sigma: [0.2, 3]
+    betag_range: [0.5, 4]
+    betap_range: [1, 2]
+
+    blur_kernel_size2: 21
+    kernel_list2: ['iso', 'aniso', 'generalized_iso', 'generalized_aniso', 'plateau_iso', 'plateau_aniso']
+    kernel_prob2: [0.45, 0.25, 0.12, 0.03, 0.12, 0.03]
+    sinc_prob2: 0.1
+    blur_sigma2: [0.2, 1.5]
+    betag_range2: [0.5, 4]
+    betap_range2: [1, 2]
+
+    final_sinc_prob: 0.8
+
+    gt_size: 256
+    use_hflip: True
+    use_rot: False
+
+    # data loader
+    use_shuffle: true
+    num_worker_per_gpu: 5
+    batch_size_per_gpu: 12
+    dataset_enlarge_ratio: 1
+    prefetch_mode: ~
+
+  # Uncomment these for validation
+  # val:
+  #   name: validation
+  #   type: PairedImageDataset
+  #   dataroot_gt: path_to_gt
+  #   dataroot_lq: path_to_lq
+  #   io_backend:
+  #     type: disk
+
+# network structures
+network_g:
+  type: RRDBNet
+  num_in_ch: 3
+  num_out_ch: 3
+  num_feat: 64
+  num_block: 23
+  num_grow_ch: 32
+
+# path
+path:
+  pretrain_network_g: experiments/pretrained_models/ESRGAN_SRx4_DF2KOST_official-ff704c30.pth
+  param_key_g: params_ema
+  strict_load_g: true
+  resume_state: ~
+
+# training settings
+train:
+  ema_decay: 0.999
+  optim_g:
+    type: Adam
+    lr: !!float 2e-4
+    weight_decay: 0
+    betas: [0.9, 0.99]
+
+  scheduler:
+    type: MultiStepLR
+    milestones: [1000000]
+    gamma: 0.5
+
+  total_iter: 1000000
+  warmup_iter: -1  # no warm up
+
+  # losses
+  pixel_opt:
+    type: L1Loss
+    loss_weight: 1.0
+    reduction: mean
+
+# Uncomment these for validation
+# validation settings
+# val:
+#   val_freq: !!float 5e3
+#   save_img: True
+
+#   metrics:
+#     psnr: # metric name
+#       type: calculate_psnr
+#       crop_border: 4
+#       test_y_channel: false
+
+# logging settings
+logger:
+  print_freq: 100
+  save_checkpoint_freq: !!float 5e3
+  use_tb_logger: true
+  wandb:
+    project: ~
+    resume_id: ~
+
+# dist training settings
+dist_params:
+  backend: nccl
+  port: 29500

Real-ESRGAN/realesrgan/__init__.py

@@ -0,0 +1,6 @@
+# flake8: noqa
+from .archs import *
+from .data import *
+from .models import *
+from .utils import *
+from .version import *

Real-ESRGAN/realesrgan/archs/__init__.py

@@ -0,0 +1,10 @@
+import importlib
+from basicsr.utils import scandir
+from os import path as osp
+
+# automatically scan and import arch modules for registry
+# scan all the files that end with '_arch.py' under the archs folder
+arch_folder = osp.dirname(osp.abspath(__file__))
+arch_filenames = [osp.splitext(osp.basename(v))[0] for v in scandir(arch_folder) if v.endswith('_arch.py')]
+# import all the arch modules
+_arch_modules = [importlib.import_module(f'realesrgan.archs.{file_name}') for file_name in arch_filenames]

Real-ESRGAN/realesrgan/archs/discriminator_arch.py

@@ -0,0 +1,67 @@
+from basicsr.utils.registry import ARCH_REGISTRY
+from torch import nn as nn
+from torch.nn import functional as F
+from torch.nn.utils import spectral_norm
+
+
+@ARCH_REGISTRY.register()
+class UNetDiscriminatorSN(nn.Module):
+    """Defines a U-Net discriminator with spectral normalization (SN)
+
+    It is used in Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data.
+
+    Arg:
+        num_in_ch (int): Channel number of inputs. Default: 3.
+        num_feat (int): Channel number of base intermediate features. Default: 64.
+        skip_connection (bool): Whether to use skip connections between U-Net. Default: True.
+    """
+
+    def __init__(self, num_in_ch, num_feat=64, skip_connection=True):
+        super(UNetDiscriminatorSN, self).__init__()
+        self.skip_connection = skip_connection
+        norm = spectral_norm
+        # the first convolution
+        self.conv0 = nn.Conv2d(num_in_ch, num_feat, kernel_size=3, stride=1, padding=1)
+        # downsample
+        self.conv1 = norm(nn.Conv2d(num_feat, num_feat * 2, 4, 2, 1, bias=False))
+        self.conv2 = norm(nn.Conv2d(num_feat * 2, num_feat * 4, 4, 2, 1, bias=False))
+        self.conv3 = norm(nn.Conv2d(num_feat * 4, num_feat * 8, 4, 2, 1, bias=False))
+        # upsample
+        self.conv4 = norm(nn.Conv2d(num_feat * 8, num_feat * 4, 3, 1, 1, bias=False))
+        self.conv5 = norm(nn.Conv2d(num_feat * 4, num_feat * 2, 3, 1, 1, bias=False))
+        self.conv6 = norm(nn.Conv2d(num_feat * 2, num_feat, 3, 1, 1, bias=False))
+        # extra convolutions
+        self.conv7 = norm(nn.Conv2d(num_feat, num_feat, 3, 1, 1, bias=False))
+        self.conv8 = norm(nn.Conv2d(num_feat, num_feat, 3, 1, 1, bias=False))
+        self.conv9 = nn.Conv2d(num_feat, 1, 3, 1, 1)
+
+    def forward(self, x):
+        # downsample
+        x0 = F.leaky_relu(self.conv0(x), negative_slope=0.2, inplace=True)
+        x1 = F.leaky_relu(self.conv1(x0), negative_slope=0.2, inplace=True)
+        x2 = F.leaky_relu(self.conv2(x1), negative_slope=0.2, inplace=True)
+        x3 = F.leaky_relu(self.conv3(x2), negative_slope=0.2, inplace=True)
+
+        # upsample
+        x3 = F.interpolate(x3, scale_factor=2, mode='bilinear', align_corners=False)
+        x4 = F.leaky_relu(self.conv4(x3), negative_slope=0.2, inplace=True)
+
+        if self.skip_connection:
+            x4 = x4 + x2
+        x4 = F.interpolate(x4, scale_factor=2, mode='bilinear', align_corners=False)
+        x5 = F.leaky_relu(self.conv5(x4), negative_slope=0.2, inplace=True)
+
+        if self.skip_connection:
+            x5 = x5 + x1
+        x5 = F.interpolate(x5, scale_factor=2, mode='bilinear', align_corners=False)
+        x6 = F.leaky_relu(self.conv6(x5), negative_slope=0.2, inplace=True)
+
+        if self.skip_connection:
+            x6 = x6 + x0
+
+        # extra convolutions
+        out = F.leaky_relu(self.conv7(x6), negative_slope=0.2, inplace=True)
+        out = F.leaky_relu(self.conv8(out), negative_slope=0.2, inplace=True)
+        out = self.conv9(out)
+
+        return out

Real-ESRGAN/realesrgan/archs/srvgg_arch.py

@@ -0,0 +1,69 @@
+from basicsr.utils.registry import ARCH_REGISTRY
+from torch import nn as nn
+from torch.nn import functional as F
+
+
+@ARCH_REGISTRY.register()
+class SRVGGNetCompact(nn.Module):
+    """A compact VGG-style network structure for super-resolution.
+
+    It is a compact network structure, which performs upsampling in the last layer and no convolution is
+    conducted on the HR feature space.
+
+    Args:
+        num_in_ch (int): Channel number of inputs. Default: 3.
+        num_out_ch (int): Channel number of outputs. Default: 3.
+        num_feat (int): Channel number of intermediate features. Default: 64.
+        num_conv (int): Number of convolution layers in the body network. Default: 16.
+        upscale (int): Upsampling factor. Default: 4.
+        act_type (str): Activation type, options: 'relu', 'prelu', 'leakyrelu'. Default: prelu.
+    """
+
+    def __init__(self, num_in_ch=3, num_out_ch=3, num_feat=64, num_conv=16, upscale=4, act_type='prelu'):
+        super(SRVGGNetCompact, self).__init__()
+        self.num_in_ch = num_in_ch
+        self.num_out_ch = num_out_ch
+        self.num_feat = num_feat
+        self.num_conv = num_conv
+        self.upscale = upscale
+        self.act_type = act_type
+
+        self.body = nn.ModuleList()
+        # the first conv
+        self.body.append(nn.Conv2d(num_in_ch, num_feat, 3, 1, 1))
+        # the first activation
+        if act_type == 'relu':
+            activation = nn.ReLU(inplace=True)
+        elif act_type == 'prelu':
+            activation = nn.PReLU(num_parameters=num_feat)
+        elif act_type == 'leakyrelu':
+            activation = nn.LeakyReLU(negative_slope=0.1, inplace=True)
+        self.body.append(activation)
+
+        # the body structure
+        for _ in range(num_conv):
+            self.body.append(nn.Conv2d(num_feat, num_feat, 3, 1, 1))
+            # activation
+            if act_type == 'relu':
+                activation = nn.ReLU(inplace=True)
+            elif act_type == 'prelu':
+                activation = nn.PReLU(num_parameters=num_feat)
+            elif act_type == 'leakyrelu':
+                activation = nn.LeakyReLU(negative_slope=0.1, inplace=True)
+            self.body.append(activation)
+
+        # the last conv
+        self.body.append(nn.Conv2d(num_feat, num_out_ch * upscale * upscale, 3, 1, 1))
+        # upsample
+        self.upsampler = nn.PixelShuffle(upscale)
+
+    def forward(self, x):
+        out = x
+        for i in range(0, len(self.body)):
+            out = self.body[i](out)
+
+        out = self.upsampler(out)
+        # add the nearest upsampled image, so that the network learns the residual
+        base = F.interpolate(x, scale_factor=self.upscale, mode='nearest')
+        out += base
+        return out

Real-ESRGAN/realesrgan/data/__init__.py

@@ -0,0 +1,10 @@
+import importlib
+from basicsr.utils import scandir
+from os import path as osp
+
+# automatically scan and import dataset modules for registry
+# scan all the files that end with '_dataset.py' under the data folder
+data_folder = osp.dirname(osp.abspath(__file__))
+dataset_filenames = [osp.splitext(osp.basename(v))[0] for v in scandir(data_folder) if v.endswith('_dataset.py')]
+# import all the dataset modules
+_dataset_modules = [importlib.import_module(f'realesrgan.data.{file_name}') for file_name in dataset_filenames]

Real-ESRGAN/realesrgan/data/realesrgan_dataset.py

@@ -0,0 +1,192 @@
+import cv2
+import math
+import numpy as np
+import os
+import os.path as osp
+import random
+import time
+import torch
+from basicsr.data.degradations import circular_lowpass_kernel, random_mixed_kernels
+from basicsr.data.transforms import augment
+from basicsr.utils import FileClient, get_root_logger, imfrombytes, img2tensor
+from basicsr.utils.registry import DATASET_REGISTRY
+from torch.utils import data as data
+
+
+@DATASET_REGISTRY.register()
+class RealESRGANDataset(data.Dataset):
+    """Dataset used for Real-ESRGAN model:
+    Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data.
+
+    It loads gt (Ground-Truth) images, and augments them.
+    It also generates blur kernels and sinc kernels for generating low-quality images.
+    Note that the low-quality images are processed in tensors on GPUS for faster processing.
+
+    Args:
+        opt (dict): Config for train datasets. It contains the following keys:
+            dataroot_gt (str): Data root path for gt.
+            meta_info (str): Path for meta information file.
+            io_backend (dict): IO backend type and other kwarg.
+            use_hflip (bool): Use horizontal flips.
+            use_rot (bool): Use rotation (use vertical flip and transposing h and w for implementation).
+            Please see more options in the codes.
+    """
+
+    def __init__(self, opt):
+        super(RealESRGANDataset, self).__init__()
+        self.opt = opt
+        self.file_client = None
+        self.io_backend_opt = opt['io_backend']
+        self.gt_folder = opt['dataroot_gt']
+
+        # file client (lmdb io backend)
+        if self.io_backend_opt['type'] == 'lmdb':
+            self.io_backend_opt['db_paths'] = [self.gt_folder]
+            self.io_backend_opt['client_keys'] = ['gt']
+            if not self.gt_folder.endswith('.lmdb'):
+                raise ValueError(f"'dataroot_gt' should end with '.lmdb', but received {self.gt_folder}")
+            with open(osp.join(self.gt_folder, 'meta_info.txt')) as fin:
+                self.paths = [line.split('.')[0] for line in fin]
+        else:
+            # disk backend with meta_info
+            # Each line in the meta_info describes the relative path to an image
+            with open(self.opt['meta_info']) as fin:
+                paths = [line.strip().split(' ')[0] for line in fin]
+                self.paths = [os.path.join(self.gt_folder, v) for v in paths]
+
+        # blur settings for the first degradation
+        self.blur_kernel_size = opt['blur_kernel_size']
+        self.kernel_list = opt['kernel_list']
+        self.kernel_prob = opt['kernel_prob']  # a list for each kernel probability
+        self.blur_sigma = opt['blur_sigma']
+        self.betag_range = opt['betag_range']  # betag used in generalized Gaussian blur kernels
+        self.betap_range = opt['betap_range']  # betap used in plateau blur kernels
+        self.sinc_prob = opt['sinc_prob']  # the probability for sinc filters
+
+        # blur settings for the second degradation
+        self.blur_kernel_size2 = opt['blur_kernel_size2']
+        self.kernel_list2 = opt['kernel_list2']
+        self.kernel_prob2 = opt['kernel_prob2']
+        self.blur_sigma2 = opt['blur_sigma2']
+        self.betag_range2 = opt['betag_range2']
+        self.betap_range2 = opt['betap_range2']
+        self.sinc_prob2 = opt['sinc_prob2']
+
+        # a final sinc filter
+        self.final_sinc_prob = opt['final_sinc_prob']
+
+        self.kernel_range = [2 * v + 1 for v in range(3, 11)]  # kernel size ranges from 7 to 21
+        # TODO: kernel range is now hard-coded, should be in the configure file
+        self.pulse_tensor = torch.zeros(21, 21).float()  # convolving with pulse tensor brings no blurry effect
+        self.pulse_tensor[10, 10] = 1
+
+    def __getitem__(self, index):
+        if self.file_client is None:
+            self.file_client = FileClient(self.io_backend_opt.pop('type'), **self.io_backend_opt)
+
+        # -------------------------------- Load gt images -------------------------------- #
+        # Shape: (h, w, c); channel order: BGR; image range: [0, 1], float32.
+        gt_path = self.paths[index]
+        # avoid errors caused by high latency in reading files
+        retry = 3
+        while retry > 0:
+            try:
+                img_bytes = self.file_client.get(gt_path, 'gt')
+            except (IOError, OSError) as e:
+                logger = get_root_logger()
+                logger.warn(f'File client error: {e}, remaining retry times: {retry - 1}')
+                # change another file to read
+                index = random.randint(0, self.__len__())
+                gt_path = self.paths[index]
+                time.sleep(1)  # sleep 1s for occasional server congestion
+            else:
+                break
+            finally:
+                retry -= 1
+        img_gt = imfrombytes(img_bytes, float32=True)
+
+        # -------------------- Do augmentation for training: flip, rotation -------------------- #
+        img_gt = augment(img_gt, self.opt['use_hflip'], self.opt['use_rot'])
+
+        # crop or pad to 400
+        # TODO: 400 is hard-coded. You may change it accordingly
+        h, w = img_gt.shape[0:2]
+        crop_pad_size = 400
+        # pad
+        if h < crop_pad_size or w < crop_pad_size:
+            pad_h = max(0, crop_pad_size - h)
+            pad_w = max(0, crop_pad_size - w)
+            img_gt = cv2.copyMakeBorder(img_gt, 0, pad_h, 0, pad_w, cv2.BORDER_REFLECT_101)
+        # crop
+        if img_gt.shape[0] > crop_pad_size or img_gt.shape[1] > crop_pad_size:
+            h, w = img_gt.shape[0:2]
+            # randomly choose top and left coordinates
+            top = random.randint(0, h - crop_pad_size)
+            left = random.randint(0, w - crop_pad_size)
+            img_gt = img_gt[top:top + crop_pad_size, left:left + crop_pad_size, ...]
+
+        # ------------------------ Generate kernels (used in the first degradation) ------------------------ #
+        kernel_size = random.choice(self.kernel_range)
+        if np.random.uniform() < self.opt['sinc_prob']:
+            # this sinc filter setting is for kernels ranging from [7, 21]
+            if kernel_size < 13:
+                omega_c = np.random.uniform(np.pi / 3, np.pi)
+            else:
+                omega_c = np.random.uniform(np.pi / 5, np.pi)
+            kernel = circular_lowpass_kernel(omega_c, kernel_size, pad_to=False)
+        else:
+            kernel = random_mixed_kernels(
+                self.kernel_list,
+                self.kernel_prob,
+                kernel_size,
+                self.blur_sigma,
+                self.blur_sigma, [-math.pi, math.pi],
+                self.betag_range,
+                self.betap_range,
+                noise_range=None)
+        # pad kernel
+        pad_size = (21 - kernel_size) // 2
+        kernel = np.pad(kernel, ((pad_size, pad_size), (pad_size, pad_size)))
+
+        # ------------------------ Generate kernels (used in the second degradation) ------------------------ #
+        kernel_size = random.choice(self.kernel_range)
+        if np.random.uniform() < self.opt['sinc_prob2']:
+            if kernel_size < 13:
+                omega_c = np.random.uniform(np.pi / 3, np.pi)
+            else:
+                omega_c = np.random.uniform(np.pi / 5, np.pi)
+            kernel2 = circular_lowpass_kernel(omega_c, kernel_size, pad_to=False)
+        else:
+            kernel2 = random_mixed_kernels(
+                self.kernel_list2,
+                self.kernel_prob2,
+                kernel_size,
+                self.blur_sigma2,
+                self.blur_sigma2, [-math.pi, math.pi],
+                self.betag_range2,
+                self.betap_range2,
+                noise_range=None)
+
+        # pad kernel
+        pad_size = (21 - kernel_size) // 2
+        kernel2 = np.pad(kernel2, ((pad_size, pad_size), (pad_size, pad_size)))
+
+        # ------------------------------------- the final sinc kernel ------------------------------------- #
+        if np.random.uniform() < self.opt['final_sinc_prob']:
+            kernel_size = random.choice(self.kernel_range)
+            omega_c = np.random.uniform(np.pi / 3, np.pi)
+            sinc_kernel = circular_lowpass_kernel(omega_c, kernel_size, pad_to=21)
+            sinc_kernel = torch.FloatTensor(sinc_kernel)
+        else:
+            sinc_kernel = self.pulse_tensor
+
+        # BGR to RGB, HWC to CHW, numpy to tensor
+        img_gt = img2tensor([img_gt], bgr2rgb=True, float32=True)[0]
+        kernel = torch.FloatTensor(kernel)
+        kernel2 = torch.FloatTensor(kernel2)
+
+        return_d = {'gt': img_gt, 'kernel1': kernel, 'kernel2': kernel2, 'sinc_kernel': sinc_kernel, 'gt_path': gt_path}
+        return return_d
+
+    def __len__(self):
+        return len(self.paths)

Real-ESRGAN/realesrgan/data/realesrgan_paired_dataset.py

@@ -0,0 +1,108 @@
+import os
+from basicsr.data.data_util import paired_paths_from_folder, paired_paths_from_lmdb
+from basicsr.data.transforms import augment, paired_random_crop
+from basicsr.utils import FileClient, imfrombytes, img2tensor
+from basicsr.utils.registry import DATASET_REGISTRY
+from torch.utils import data as data
+from torchvision.transforms.functional import normalize
+
+
+@DATASET_REGISTRY.register()
+class RealESRGANPairedDataset(data.Dataset):
+    """Paired image dataset for image restoration.
+
+    Read LQ (Low Quality, e.g. LR (Low Resolution), blurry, noisy, etc) and GT image pairs.
+
+    There are three modes:
+    1. 'lmdb': Use lmdb files.
+        If opt['io_backend'] == lmdb.
+    2. 'meta_info': Use meta information file to generate paths.
+        If opt['io_backend'] != lmdb and opt['meta_info'] is not None.
+    3. 'folder': Scan folders to generate paths.
+        The rest.
+
+    Args:
+        opt (dict): Config for train datasets. It contains the following keys:
+            dataroot_gt (str): Data root path for gt.
+            dataroot_lq (str): Data root path for lq.
+            meta_info (str): Path for meta information file.
+            io_backend (dict): IO backend type and other kwarg.
+            filename_tmpl (str): Template for each filename. Note that the template excludes the file extension.
+                Default: '{}'.
+            gt_size (int): Cropped patched size for gt patches.
+            use_hflip (bool): Use horizontal flips.
+            use_rot (bool): Use rotation (use vertical flip and transposing h
+                and w for implementation).
+
+            scale (bool): Scale, which will be added automatically.
+            phase (str): 'train' or 'val'.
+    """
+
+    def __init__(self, opt):
+        super(RealESRGANPairedDataset, self).__init__()
+        self.opt = opt
+        self.file_client = None
+        self.io_backend_opt = opt['io_backend']
+        # mean and std for normalizing the input images
+        self.mean = opt['mean'] if 'mean' in opt else None
+        self.std = opt['std'] if 'std' in opt else None
+
+        self.gt_folder, self.lq_folder = opt['dataroot_gt'], opt['dataroot_lq']
+        self.filename_tmpl = opt['filename_tmpl'] if 'filename_tmpl' in opt else '{}'
+
+        # file client (lmdb io backend)
+        if self.io_backend_opt['type'] == 'lmdb':
+            self.io_backend_opt['db_paths'] = [self.lq_folder, self.gt_folder]
+            self.io_backend_opt['client_keys'] = ['lq', 'gt']
+            self.paths = paired_paths_from_lmdb([self.lq_folder, self.gt_folder], ['lq', 'gt'])
+        elif 'meta_info' in self.opt and self.opt['meta_info'] is not None:
+            # disk backend with meta_info
+            # Each line in the meta_info describes the relative path to an image
+            with open(self.opt['meta_info']) as fin:
+                paths = [line.strip() for line in fin]
+            self.paths = []
+            for path in paths:
+                gt_path, lq_path = path.split(', ')
+                gt_path = os.path.join(self.gt_folder, gt_path)
+                lq_path = os.path.join(self.lq_folder, lq_path)
+                self.paths.append(dict([('gt_path', gt_path), ('lq_path', lq_path)]))
+        else:
+            # disk backend
+            # it will scan the whole folder to get meta info
+            # it will be time-consuming for folders with too many files. It is recommended using an extra meta txt file
+            self.paths = paired_paths_from_folder([self.lq_folder, self.gt_folder], ['lq', 'gt'], self.filename_tmpl)
+
+    def __getitem__(self, index):
+        if self.file_client is None:
+            self.file_client = FileClient(self.io_backend_opt.pop('type'), **self.io_backend_opt)
+
+        scale = self.opt['scale']
+
+        # Load gt and lq images. Dimension order: HWC; channel order: BGR;
+        # image range: [0, 1], float32.
+        gt_path = self.paths[index]['gt_path']
+        img_bytes = self.file_client.get(gt_path, 'gt')
+        img_gt = imfrombytes(img_bytes, float32=True)
+        lq_path = self.paths[index]['lq_path']
+        img_bytes = self.file_client.get(lq_path, 'lq')
+        img_lq = imfrombytes(img_bytes, float32=True)
+
+        # augmentation for training
+        if self.opt['phase'] == 'train':
+            gt_size = self.opt['gt_size']
+            # random crop
+            img_gt, img_lq = paired_random_crop(img_gt, img_lq, gt_size, scale, gt_path)
+            # flip, rotation
+            img_gt, img_lq = augment([img_gt, img_lq], self.opt['use_hflip'], self.opt['use_rot'])
+
+        # BGR to RGB, HWC to CHW, numpy to tensor
+        img_gt, img_lq = img2tensor([img_gt, img_lq], bgr2rgb=True, float32=True)
+        # normalize
+        if self.mean is not None or self.std is not None:
+            normalize(img_lq, self.mean, self.std, inplace=True)
+            normalize(img_gt, self.mean, self.std, inplace=True)
+
+        return {'lq': img_lq, 'gt': img_gt, 'lq_path': lq_path, 'gt_path': gt_path}
+
+    def __len__(self):
+        return len(self.paths)

Real-ESRGAN/realesrgan/models/__init__.py

@@ -0,0 +1,10 @@
+import importlib
+from basicsr.utils import scandir
+from os import path as osp
+
+# automatically scan and import model modules for registry
+# scan all the files that end with '_model.py' under the model folder
+model_folder = osp.dirname(osp.abspath(__file__))
+model_filenames = [osp.splitext(osp.basename(v))[0] for v in scandir(model_folder) if v.endswith('_model.py')]
+# import all the model modules
+_model_modules = [importlib.import_module(f'realesrgan.models.{file_name}') for file_name in model_filenames]

Real-ESRGAN/realesrgan/models/realesrgan_model.py

@@ -0,0 +1,258 @@
+import numpy as np
+import random
+import torch
+from basicsr.data.degradations import random_add_gaussian_noise_pt, random_add_poisson_noise_pt
+from basicsr.data.transforms import paired_random_crop
+from basicsr.models.srgan_model import SRGANModel
+from basicsr.utils import DiffJPEG, USMSharp
+from basicsr.utils.img_process_util import filter2D
+from basicsr.utils.registry import MODEL_REGISTRY
+from collections import OrderedDict
+from torch.nn import functional as F
+
+
+@MODEL_REGISTRY.register()
+class RealESRGANModel(SRGANModel):
+    """RealESRGAN Model for Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data.
+
+    It mainly performs:
+    1. randomly synthesize LQ images in GPU tensors
+    2. optimize the networks with GAN training.
+    """
+
+    def __init__(self, opt):
+        super(RealESRGANModel, self).__init__(opt)
+        self.jpeger = DiffJPEG(differentiable=False).cuda()  # simulate JPEG compression artifacts
+        self.usm_sharpener = USMSharp().cuda()  # do usm sharpening
+        self.queue_size = opt.get('queue_size', 180)
+
+    @torch.no_grad()
+    def _dequeue_and_enqueue(self):
+        """It is the training pair pool for increasing the diversity in a batch.
+
+        Batch processing limits the diversity of synthetic degradations in a batch. For example, samples in a
+        batch could not have different resize scaling factors. Therefore, we employ this training pair pool
+        to increase the degradation diversity in a batch.
+        """
+        # initialize
+        b, c, h, w = self.lq.size()
+        if not hasattr(self, 'queue_lr'):
+            assert self.queue_size % b == 0, f'queue size {self.queue_size} should be divisible by batch size {b}'
+            self.queue_lr = torch.zeros(self.queue_size, c, h, w).cuda()
+            _, c, h, w = self.gt.size()
+            self.queue_gt = torch.zeros(self.queue_size, c, h, w).cuda()
+            self.queue_ptr = 0
+        if self.queue_ptr == self.queue_size:  # the pool is full
+            # do dequeue and enqueue
+            # shuffle
+            idx = torch.randperm(self.queue_size)
+            self.queue_lr = self.queue_lr[idx]
+            self.queue_gt = self.queue_gt[idx]
+            # get first b samples
+            lq_dequeue = self.queue_lr[0:b, :, :, :].clone()
+            gt_dequeue = self.queue_gt[0:b, :, :, :].clone()
+            # update the queue
+            self.queue_lr[0:b, :, :, :] = self.lq.clone()
+            self.queue_gt[0:b, :, :, :] = self.gt.clone()
+
+            self.lq = lq_dequeue
+            self.gt = gt_dequeue
+        else:
+            # only do enqueue
+            self.queue_lr[self.queue_ptr:self.queue_ptr + b, :, :, :] = self.lq.clone()
+            self.queue_gt[self.queue_ptr:self.queue_ptr + b, :, :, :] = self.gt.clone()
+            self.queue_ptr = self.queue_ptr + b
+
+    @torch.no_grad()
+    def feed_data(self, data):
+        """Accept data from dataloader, and then add two-order degradations to obtain LQ images.
+        """
+        if self.is_train and self.opt.get('high_order_degradation', True):
+            # training data synthesis
+            self.gt = data['gt'].to(self.device)
+            self.gt_usm = self.usm_sharpener(self.gt)
+
+            self.kernel1 = data['kernel1'].to(self.device)
+            self.kernel2 = data['kernel2'].to(self.device)
+            self.sinc_kernel = data['sinc_kernel'].to(self.device)
+
+            ori_h, ori_w = self.gt.size()[2:4]
+
+            # ----------------------- The first degradation process ----------------------- #
+            # blur
+            out = filter2D(self.gt_usm, self.kernel1)
+            # random resize
+            updown_type = random.choices(['up', 'down', 'keep'], self.opt['resize_prob'])[0]
+            if updown_type == 'up':
+                scale = np.random.uniform(1, self.opt['resize_range'][1])
+            elif updown_type == 'down':
+                scale = np.random.uniform(self.opt['resize_range'][0], 1)
+            else:
+                scale = 1
+            mode = random.choice(['area', 'bilinear', 'bicubic'])
+            out = F.interpolate(out, scale_factor=scale, mode=mode)
+            # add noise
+            gray_noise_prob = self.opt['gray_noise_prob']
+            if np.random.uniform() < self.opt['gaussian_noise_prob']:
+                out = random_add_gaussian_noise_pt(
+                    out, sigma_range=self.opt['noise_range'], clip=True, rounds=False, gray_prob=gray_noise_prob)
+            else:
+                out = random_add_poisson_noise_pt(
+                    out,
+                    scale_range=self.opt['poisson_scale_range'],
+                    gray_prob=gray_noise_prob,
+                    clip=True,
+                    rounds=False)
+            # JPEG compression
+            jpeg_p = out.new_zeros(out.size(0)).uniform_(*self.opt['jpeg_range'])
+            out = torch.clamp(out, 0, 1)  # clamp to [0, 1], otherwise JPEGer will result in unpleasant artifacts
+            out = self.jpeger(out, quality=jpeg_p)
+
+            # ----------------------- The second degradation process ----------------------- #
+            # blur
+            if np.random.uniform() < self.opt['second_blur_prob']:
+                out = filter2D(out, self.kernel2)
+            # random resize
+            updown_type = random.choices(['up', 'down', 'keep'], self.opt['resize_prob2'])[0]
+            if updown_type == 'up':
+                scale = np.random.uniform(1, self.opt['resize_range2'][1])
+            elif updown_type == 'down':
+                scale = np.random.uniform(self.opt['resize_range2'][0], 1)
+            else:
+                scale = 1
+            mode = random.choice(['area', 'bilinear', 'bicubic'])
+            out = F.interpolate(
+                out, size=(int(ori_h / self.opt['scale'] * scale), int(ori_w / self.opt['scale'] * scale)), mode=mode)
+            # add noise
+            gray_noise_prob = self.opt['gray_noise_prob2']
+            if np.random.uniform() < self.opt['gaussian_noise_prob2']:
+                out = random_add_gaussian_noise_pt(
+                    out, sigma_range=self.opt['noise_range2'], clip=True, rounds=False, gray_prob=gray_noise_prob)
+            else:
+                out = random_add_poisson_noise_pt(
+                    out,
+                    scale_range=self.opt['poisson_scale_range2'],
+                    gray_prob=gray_noise_prob,
+                    clip=True,
+                    rounds=False)
+
+            # JPEG compression + the final sinc filter
+            # We also need to resize images to desired sizes. We group [resize back + sinc filter] together
+            # as one operation.
+            # We consider two orders:
+            #   1. [resize back + sinc filter] + JPEG compression
+            #   2. JPEG compression + [resize back + sinc filter]
+            # Empirically, we find other combinations (sinc + JPEG + Resize) will introduce twisted lines.
+            if np.random.uniform() < 0.5:
+                # resize back + the final sinc filter
+                mode = random.choice(['area', 'bilinear', 'bicubic'])
+                out = F.interpolate(out, size=(ori_h // self.opt['scale'], ori_w // self.opt['scale']), mode=mode)
+                out = filter2D(out, self.sinc_kernel)
+                # JPEG compression
+                jpeg_p = out.new_zeros(out.size(0)).uniform_(*self.opt['jpeg_range2'])
+                out = torch.clamp(out, 0, 1)
+                out = self.jpeger(out, quality=jpeg_p)
+            else:
+                # JPEG compression
+                jpeg_p = out.new_zeros(out.size(0)).uniform_(*self.opt['jpeg_range2'])
+                out = torch.clamp(out, 0, 1)
+                out = self.jpeger(out, quality=jpeg_p)
+                # resize back + the final sinc filter
+                mode = random.choice(['area', 'bilinear', 'bicubic'])
+                out = F.interpolate(out, size=(ori_h // self.opt['scale'], ori_w // self.opt['scale']), mode=mode)
+                out = filter2D(out, self.sinc_kernel)
+
+            # clamp and round
+            self.lq = torch.clamp((out * 255.0).round(), 0, 255) / 255.
+
+            # random crop
+            gt_size = self.opt['gt_size']
+            (self.gt, self.gt_usm), self.lq = paired_random_crop([self.gt, self.gt_usm], self.lq, gt_size,
+                                                                 self.opt['scale'])
+
+            # training pair pool
+            self._dequeue_and_enqueue()
+            # sharpen self.gt again, as we have changed the self.gt with self._dequeue_and_enqueue
+            self.gt_usm = self.usm_sharpener(self.gt)
+            self.lq = self.lq.contiguous()  # for the warning: grad and param do not obey the gradient layout contract
+        else:
+            # for paired training or validation
+            self.lq = data['lq'].to(self.device)
+            if 'gt' in data:
+                self.gt = data['gt'].to(self.device)
+                self.gt_usm = self.usm_sharpener(self.gt)
+
+    def nondist_validation(self, dataloader, current_iter, tb_logger, save_img):
+        # do not use the synthetic process during validation
+        self.is_train = False
+        super(RealESRGANModel, self).nondist_validation(dataloader, current_iter, tb_logger, save_img)
+        self.is_train = True
+
+    def optimize_parameters(self, current_iter):
+        # usm sharpening
+        l1_gt = self.gt_usm
+        percep_gt = self.gt_usm
+        gan_gt = self.gt_usm
+        if self.opt['l1_gt_usm'] is False:
+            l1_gt = self.gt
+        if self.opt['percep_gt_usm'] is False:
+            percep_gt = self.gt
+        if self.opt['gan_gt_usm'] is False:
+            gan_gt = self.gt
+
+        # optimize net_g
+        for p in self.net_d.parameters():
+            p.requires_grad = False
+
+        self.optimizer_g.zero_grad()
+        self.output = self.net_g(self.lq)
+
+        l_g_total = 0
+        loss_dict = OrderedDict()
+        if (current_iter % self.net_d_iters == 0 and current_iter > self.net_d_init_iters):
+            # pixel loss
+            if self.cri_pix:
+                l_g_pix = self.cri_pix(self.output, l1_gt)
+                l_g_total += l_g_pix
+                loss_dict['l_g_pix'] = l_g_pix
+            # perceptual loss
+            if self.cri_perceptual:
+                l_g_percep, l_g_style = self.cri_perceptual(self.output, percep_gt)
+                if l_g_percep is not None:
+                    l_g_total += l_g_percep
+                    loss_dict['l_g_percep'] = l_g_percep
+                if l_g_style is not None:
+                    l_g_total += l_g_style
+                    loss_dict['l_g_style'] = l_g_style
+            # gan loss
+            fake_g_pred = self.net_d(self.output)
+            l_g_gan = self.cri_gan(fake_g_pred, True, is_disc=False)
+            l_g_total += l_g_gan
+            loss_dict['l_g_gan'] = l_g_gan
+
+            l_g_total.backward()
+            self.optimizer_g.step()
+
+        # optimize net_d
+        for p in self.net_d.parameters():
+            p.requires_grad = True
+
+        self.optimizer_d.zero_grad()
+        # real
+        real_d_pred = self.net_d(gan_gt)
+        l_d_real = self.cri_gan(real_d_pred, True, is_disc=True)
+        loss_dict['l_d_real'] = l_d_real
+        loss_dict['out_d_real'] = torch.mean(real_d_pred.detach())
+        l_d_real.backward()
+        # fake
+        fake_d_pred = self.net_d(self.output.detach().clone())  # clone for pt1.9
+        l_d_fake = self.cri_gan(fake_d_pred, False, is_disc=True)
+        loss_dict['l_d_fake'] = l_d_fake
+        loss_dict['out_d_fake'] = torch.mean(fake_d_pred.detach())
+        l_d_fake.backward()
+        self.optimizer_d.step()
+
+        if self.ema_decay > 0:
+            self.model_ema(decay=self.ema_decay)
+
+        self.log_dict = self.reduce_loss_dict(loss_dict)

Real-ESRGAN/realesrgan/models/realesrnet_model.py

@@ -0,0 +1,188 @@
+import numpy as np
+import random
+import torch
+from basicsr.data.degradations import random_add_gaussian_noise_pt, random_add_poisson_noise_pt
+from basicsr.data.transforms import paired_random_crop
+from basicsr.models.sr_model import SRModel
+from basicsr.utils import DiffJPEG, USMSharp
+from basicsr.utils.img_process_util import filter2D
+from basicsr.utils.registry import MODEL_REGISTRY
+from torch.nn import functional as F
+
+
+@MODEL_REGISTRY.register()
+class RealESRNetModel(SRModel):
+    """RealESRNet Model for Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data.
+
+    It is trained without GAN losses.
+    It mainly performs:
+    1. randomly synthesize LQ images in GPU tensors
+    2. optimize the networks with GAN training.
+    """
+
+    def __init__(self, opt):
+        super(RealESRNetModel, self).__init__(opt)
+        self.jpeger = DiffJPEG(differentiable=False).cuda()  # simulate JPEG compression artifacts
+        self.usm_sharpener = USMSharp().cuda()  # do usm sharpening
+        self.queue_size = opt.get('queue_size', 180)
+
+    @torch.no_grad()
+    def _dequeue_and_enqueue(self):
+        """It is the training pair pool for increasing the diversity in a batch.
+
+        Batch processing limits the diversity of synthetic degradations in a batch. For example, samples in a
+        batch could not have different resize scaling factors. Therefore, we employ this training pair pool
+        to increase the degradation diversity in a batch.
+        """
+        # initialize
+        b, c, h, w = self.lq.size()
+        if not hasattr(self, 'queue_lr'):
+            assert self.queue_size % b == 0, f'queue size {self.queue_size} should be divisible by batch size {b}'
+            self.queue_lr = torch.zeros(self.queue_size, c, h, w).cuda()
+            _, c, h, w = self.gt.size()
+            self.queue_gt = torch.zeros(self.queue_size, c, h, w).cuda()
+            self.queue_ptr = 0
+        if self.queue_ptr == self.queue_size:  # the pool is full
+            # do dequeue and enqueue
+            # shuffle
+            idx = torch.randperm(self.queue_size)
+            self.queue_lr = self.queue_lr[idx]
+            self.queue_gt = self.queue_gt[idx]
+            # get first b samples
+            lq_dequeue = self.queue_lr[0:b, :, :, :].clone()
+            gt_dequeue = self.queue_gt[0:b, :, :, :].clone()
+            # update the queue
+            self.queue_lr[0:b, :, :, :] = self.lq.clone()
+            self.queue_gt[0:b, :, :, :] = self.gt.clone()
+
+            self.lq = lq_dequeue
+            self.gt = gt_dequeue
+        else:
+            # only do enqueue
+            self.queue_lr[self.queue_ptr:self.queue_ptr + b, :, :, :] = self.lq.clone()
+            self.queue_gt[self.queue_ptr:self.queue_ptr + b, :, :, :] = self.gt.clone()
+            self.queue_ptr = self.queue_ptr + b
+
+    @torch.no_grad()
+    def feed_data(self, data):
+        """Accept data from dataloader, and then add two-order degradations to obtain LQ images.
+        """
+        if self.is_train and self.opt.get('high_order_degradation', True):
+            # training data synthesis
+            self.gt = data['gt'].to(self.device)
+            # USM sharpen the GT images
+            if self.opt['gt_usm'] is True:
+                self.gt = self.usm_sharpener(self.gt)
+
+            self.kernel1 = data['kernel1'].to(self.device)
+            self.kernel2 = data['kernel2'].to(self.device)
+            self.sinc_kernel = data['sinc_kernel'].to(self.device)
+
+            ori_h, ori_w = self.gt.size()[2:4]
+
+            # ----------------------- The first degradation process ----------------------- #
+            # blur
+            out = filter2D(self.gt, self.kernel1)
+            # random resize
+            updown_type = random.choices(['up', 'down', 'keep'], self.opt['resize_prob'])[0]
+            if updown_type == 'up':
+                scale = np.random.uniform(1, self.opt['resize_range'][1])
+            elif updown_type == 'down':
+                scale = np.random.uniform(self.opt['resize_range'][0], 1)
+            else:
+                scale = 1
+            mode = random.choice(['area', 'bilinear', 'bicubic'])
+            out = F.interpolate(out, scale_factor=scale, mode=mode)
+            # add noise
+            gray_noise_prob = self.opt['gray_noise_prob']
+            if np.random.uniform() < self.opt['gaussian_noise_prob']:
+                out = random_add_gaussian_noise_pt(
+                    out, sigma_range=self.opt['noise_range'], clip=True, rounds=False, gray_prob=gray_noise_prob)
+            else:
+                out = random_add_poisson_noise_pt(
+                    out,
+                    scale_range=self.opt['poisson_scale_range'],
+                    gray_prob=gray_noise_prob,
+                    clip=True,
+                    rounds=False)
+            # JPEG compression
+            jpeg_p = out.new_zeros(out.size(0)).uniform_(*self.opt['jpeg_range'])
+            out = torch.clamp(out, 0, 1)  # clamp to [0, 1], otherwise JPEGer will result in unpleasant artifacts
+            out = self.jpeger(out, quality=jpeg_p)
+
+            # ----------------------- The second degradation process ----------------------- #
+            # blur
+            if np.random.uniform() < self.opt['second_blur_prob']:
+                out = filter2D(out, self.kernel2)
+            # random resize
+            updown_type = random.choices(['up', 'down', 'keep'], self.opt['resize_prob2'])[0]
+            if updown_type == 'up':
+                scale = np.random.uniform(1, self.opt['resize_range2'][1])
+            elif updown_type == 'down':
+                scale = np.random.uniform(self.opt['resize_range2'][0], 1)
+            else:
+                scale = 1
+            mode = random.choice(['area', 'bilinear', 'bicubic'])
+            out = F.interpolate(
+                out, size=(int(ori_h / self.opt['scale'] * scale), int(ori_w / self.opt['scale'] * scale)), mode=mode)
+            # add noise
+            gray_noise_prob = self.opt['gray_noise_prob2']
+            if np.random.uniform() < self.opt['gaussian_noise_prob2']:
+                out = random_add_gaussian_noise_pt(
+                    out, sigma_range=self.opt['noise_range2'], clip=True, rounds=False, gray_prob=gray_noise_prob)
+            else:
+                out = random_add_poisson_noise_pt(
+                    out,
+                    scale_range=self.opt['poisson_scale_range2'],
+                    gray_prob=gray_noise_prob,
+                    clip=True,
+                    rounds=False)
+
+            # JPEG compression + the final sinc filter
+            # We also need to resize images to desired sizes. We group [resize back + sinc filter] together
+            # as one operation.
+            # We consider two orders:
+            #   1. [resize back + sinc filter] + JPEG compression
+            #   2. JPEG compression + [resize back + sinc filter]
+            # Empirically, we find other combinations (sinc + JPEG + Resize) will introduce twisted lines.
+            if np.random.uniform() < 0.5:
+                # resize back + the final sinc filter
+                mode = random.choice(['area', 'bilinear', 'bicubic'])
+                out = F.interpolate(out, size=(ori_h // self.opt['scale'], ori_w // self.opt['scale']), mode=mode)
+                out = filter2D(out, self.sinc_kernel)
+                # JPEG compression
+                jpeg_p = out.new_zeros(out.size(0)).uniform_(*self.opt['jpeg_range2'])
+                out = torch.clamp(out, 0, 1)
+                out = self.jpeger(out, quality=jpeg_p)
+            else:
+                # JPEG compression
+                jpeg_p = out.new_zeros(out.size(0)).uniform_(*self.opt['jpeg_range2'])
+                out = torch.clamp(out, 0, 1)
+                out = self.jpeger(out, quality=jpeg_p)
+                # resize back + the final sinc filter
+                mode = random.choice(['area', 'bilinear', 'bicubic'])
+                out = F.interpolate(out, size=(ori_h // self.opt['scale'], ori_w // self.opt['scale']), mode=mode)
+                out = filter2D(out, self.sinc_kernel)
+
+            # clamp and round
+            self.lq = torch.clamp((out * 255.0).round(), 0, 255) / 255.
+
+            # random crop
+            gt_size = self.opt['gt_size']
+            self.gt, self.lq = paired_random_crop(self.gt, self.lq, gt_size, self.opt['scale'])
+
+            # training pair pool
+            self._dequeue_and_enqueue()
+            self.lq = self.lq.contiguous()  # for the warning: grad and param do not obey the gradient layout contract
+        else:
+            # for paired training or validation
+            self.lq = data['lq'].to(self.device)
+            if 'gt' in data:
+                self.gt = data['gt'].to(self.device)
+                self.gt_usm = self.usm_sharpener(self.gt)
+
+    def nondist_validation(self, dataloader, current_iter, tb_logger, save_img):
+        # do not use the synthetic process during validation
+        self.is_train = False
+        super(RealESRNetModel, self).nondist_validation(dataloader, current_iter, tb_logger, save_img)
+        self.is_train = True