【论文复现】DCFace: Synthetic Face Generation with Dual Condition Diffusion Model

DCFace: 用双条件扩散模型生成合成人脸。CVPR 2023.4

code：mk-minchul/dcface (github.com)

paper：[2304.07060] DCFace: Synthetic Face Generation with Dual Condition Diffusion Model (arxiv.org)

论文

介绍

应用：生成用于训练人脸识别模型的合成数据集。

背景：生成合成数据集需要考虑多个因素，如姿势、光照、表情、年龄和遮挡等，以模拟真实图像的条件分布。以往的研究主要使用GAN或3D模型生成合成数据集。

方法：本文提出一种基于扩散模型的双重条件人脸生成器（DCFace），通过结合主体外观（ID）和外部因素（风格）条件来控制类内和类间变化。作者使用了新颖的Patch-wise风格提取器和Time-step依赖的ID损失，使DCFace能够在不同风格下精确控制同一主体的人脸图像生成。

结果：使用DCFace生成的合成数据集训练的人脸识别模型在LFW、CFP-FP、CPLFW、AgeDB和CALFW等5个测试数据集中，平均比以前的方法提高了6.11％的验证准确性。

方法

 

实验

 

 

 

 

 

结论

复现

Image generation

问题1：No models "dcface/dcface/pretrained_models/adaface_ir101_webface4m.ckpt'

Traceback (most recent call last):
  File "/data/dcface/dcface/src/recognition/recognition_helper.py", line 53, in download_ir_pretrained_statedict
    subprocess.check_call([os.path.expanduser('/opt/conda/envs/dcface/lib/python3.7/site-packages/gdown'), '--id', _id])
  File "/opt/conda/envs/dcface/lib/python3.7/subprocess.py", line 358, in check_call
    retcode = call(*popenargs, **kwargs)
  File "/opt/conda/envs/dcface/lib/python3.7/subprocess.py", line 339, in call
    with Popen(*popenargs, **kwargs) as p:
  File "/opt/conda/envs/dcface/lib/python3.7/subprocess.py", line 800, in __init__
    restore_signals, start_new_session)
  File "/opt/conda/envs/dcface/lib/python3.7/subprocess.py", line 1551, in _execute_child
    raise child_exception_type(errno_num, err_msg, err_filename)
PermissionError: [Errno 13] Permission denied: '/opt/conda/envs/dcface/lib/python3.7/site-packages/gdown'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "synthesis.py", line 140, in 
    main()
  File "synthesis.py", line 77, in main
    pl_module: LightningModule = hydra.utils.instantiate(model_hparam)()
  File "/data/dcface/dcface/src/trainer.py", line 57, in __init__
    self.recognition_model_eval: RecognitionModel = make_recognition_model(self.hparams.recognition_eval)
  File "/data/dcface/dcface/src/recognition/recognition_helper.py", line 264, in make_recognition_model
    model_statedict = download_ir_pretrained_statedict(backbone_name, 'webface4m', 'adaface')
  File "/data/dcface/dcface/src/recognition/recognition_helper.py", line 56, in download_ir_pretrained_statedict
    subprocess.check_call([os.path.expanduser('/opt/conda/envs/dcface/lib/python3.7/site-packages/gdown'), '--id', _id])
  File "/opt/conda/envs/dcface/lib/python3.7/subprocess.py", line 358, in check_call
    retcode = call(*popenargs, **kwargs)
  File "/opt/conda/envs/dcface/lib/python3.7/subprocess.py", line 339, in call
    with Popen(*popenargs, **kwargs) as p:
  File "/opt/conda/envs/dcface/lib/python3.7/subprocess.py", line 800, in __init__
    restore_signals, start_new_session)
  File "/opt/conda/envs/dcface/lib/python3.7/subprocess.py", line 1551, in _execute_child
    raise child_exception_type(errno_num, err_msg, err_filename)
PermissionError: [Errno 13] Permission denied: '/opt/conda/envs/dcface/lib/python3.7/site-packages/gdown'

 报错没直接报缺少文件，而是一个包gdown的问题，解决的方向就错了，print了一下发现是因为checkpoint_path的文件缺失，发现缺失的权重作者没给出，搜索后在作者的另一个项目里找到了：

参考：mk-minchul/AdaFace (github.com)

Google 云端硬盘 - 病毒扫描警告

 id_image:

style_image:

 

generated_images:

 

generate dataset 

Training generator

Training Face Recognition Model

to be released 

dcface datasets:https://github.com/mk-minchul/dcface?tab=readme-ov-file#dataset-release