Code Monkey home page Code Monkey logo

retouching_ffhq_detection's Introduction

RetouchingFFHQ-A-Large-scale-Dataset-for-Fine-grained-Face-Retouching-Detection

MM23: RetouchingFFHQ: A Large-scale Dataset for Fine-grained Face Retouching Detection

Hopefully this dataset & work would benefit you.

Paper: https://dl.acm.org/doi/pdf/10.1145/3581783.3611843

Dataset request: https://fdmas.github.io/Application_RetouchingFFHQ_new.pdf

checkpoints: https://drive.google.com/drive/folders/1BoVqkIK33bmwXTuBTXLT6zLjwpbOgAQW?usp=sharing (the files are renamed upon upload. please refer to meiyan_ckpt2yml.txt)

Should you encounter problem, contact me via [email protected], or open issues. Thanks!

为了方便阅读,本说明文件尽量用中文来表示

文件架构介绍

  • data: 这里存放的是数据集的读取方法以及dataloader
  • (+)model:这里存放的是模型脚本,其中base_model.py是基类,定义了一些基础方法,比如对图像进行攻击、模型保存等等。meiyan_baseline.py是对基类方法的继承,定义具体实验的train/test函数等。networks.py定义了常用网络架构
  • noise_layers:定义了一些图像后处理攻击,例如blur、JPEG压缩等等
  • (+)options:这里存放的是配置文件和配置文件如何读取,options.py是定义怎么读取对应的yml
  • utils:其他的一些杂七杂八的帮助函数,比如创建文件夹之类的
  • (+)sh: 运行的脚本
  • (+)train.py,运行的主入口

控制台输出

控制台会收集模型train/test过程中的变量并自动求平均,把需要统计的变量(meiyan_baseline.py的Line 82/83塞了acc和loss)放进logs的字典里返回即可

求平均的逻辑在 Line 110-130

以下是控制台的样例

[1, 5000 79968 1 0.0001] lr: 0.0001 epoch_accuracy: 0.1296 loss: 2.3517 time per sample 0.0040 s
[1, 5000 79968 0 0.0001] lr: 0.0001 epoch_accuracy: 0.1291 loss: 2.3527 time per sample 0.0040 s
Saving models and training states.
Model saved to: /groupshare/meiyan_detection_results/model/ViT/4999_ViT.pth
[1, 5040 80608 1 0.0001] lr: 0.0001 epoch_accuracy: 0.1296 loss: 2.3515 time per sample 0.0040 s
[1, 5040 80608 0 0.0001] lr: 0.0001 epoch_accuracy: 0.1292 loss: 2.3525 time per sample 0.0040 s

模型与结果保存路径

模型和图像结果没有保存在项目文件夹下面,这个很重要,因为整个文件夹Git了,如果存在里面的话会导致Git的东西很多 已经设置好路径,为/groupshare/meiyan_detection_results,这个的设置在 Line 61, base_model.py (self.out_space_storage) 在sh里面可以指定task_name,这样的话不同实验可以有单独的二级目录,比如如果task_name="ViT",则模型会在/groupshare/meiyan_detection_results/ViT/models下面

工作流

当你调用了sh文件(例如run_detection.sh)后:

python -m torch.distributed.launch --master_port 3111 --nproc_per_node=2 train.py \
          -opt options/meiyan_hallucinate.yml -mode 0 --launcher pytorch
  • 参数: -opt会指定配置文件使用哪个,-task_name指定本次实验的名字(上面说过了,用来存模型指定路径等等),-nproc_per_node是枝使用几张卡,这个要记得和yml里面的gpu_ids(例如meiyan_hallucinate.yml的Line 7)保持一致
  • 数据集加载: train.py Line 88 (这一行上面的都是定义args和启动分布式训练的语句,不用怎么修改)
  • 模型定义: train.py Line 92
  • 训练/测试脚本:train.py Line 98
  • 标准训练测试脚本分为两个步骤:数据读入(例:feed_data_router)以及执行(例:train_ViT)

retouching_ffhq_detection's People

Contributors

yingqichao avatar

Stargazers

 avatar  avatar  avatar

Watchers

 avatar

retouching_ffhq_detection's Issues

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.