Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

how to use syncnet_python and training steps #124

Open
1129571 opened this issue Feb 19, 2024 · 11 comments
Open

how to use syncnet_python and training steps #124

1129571 opened this issue Feb 19, 2024 · 11 comments

Comments

@1129571
Copy link

1129571 commented Feb 19, 2024

  1. 25fps, 16000Hz, less than 5s
  2. Run run_pipeline.py (in syncnet_python)
  3. Run run_syncnet.py. (in syncnet_python) Then determine the value of AV offset. If it is between [-1,1], keep the video. Otherwise, remove the video from the dataset?
  4. Run preprocess.py
  5. Run train_syncnet_sam.py
  6. Run wav2lip_train.py

Is this process correct?
Especially, I want to know if the third step is correct!

@tailangjun
Copy link

tailangjun commented Feb 19, 2024

  1. 25fps, 16000Hz, less than 5s
  2. Run run_pipeline.py (in syncnet_python)
  3. Run run_syncnet.py. (in syncnet_python) Then determine the value of AV offset. If it is between [-1,1], keep the video. Otherwise, remove the video from the dataset?
  4. Run preprocess.py
  5. Run train_syncnet_sam.py
  6. Run wav2lip_train.py

Is this process correct? Especially, I want to know if the third step is correct!

代码没问题的,你可以参考本工程的 ./evaluation/scores_LSE/calculate_scores_real_videos
重点是 offset, conf, dist = s.evaluate(opt,videofile=fname),你需要对 offset和 conf添加过滤条件

@1129571
Copy link
Author

1129571 commented Feb 19, 2024

  1. 25fps, 16000Hz, less than 5s
  2. Run run_pipeline.py (in syncnet_python)
  3. Run run_syncnet.py. (in syncnet_python) Then determine the value of AV offset. If it is between [-1,1], keep the video. Otherwise, remove the video from the dataset?
  4. Run preprocess.py
  5. Run train_syncnet_sam.py
  6. Run wav2lip_train.py

Is this process correct? Especially, I want to know if the third step is correct!

代码没问题的,你可以参考本工程的 ./evaluation/scores_LSE/calculate_scores_real_videos 重点是 offset, conf, dist = s.evaluate(opt,videofile=fname),你需要对 offset和 conf添加过滤条件
兄弟可以详细一点吗,我没太明白。我正在用lrs2在进行训练,但是loss一直在0.69,issues里说syncnet_python可以解决。
./evaluation/scores_LSE/calculate_scores_real_videos和syncnet_python相似?评估数据。
正确做法是将offset和confidence评估不符合的数据从数据集移除?(还是说怎么样操作)

@tailangjun
Copy link

将offset和confidence评估不符合的数据从数据集移除?

是的,将offset和confidence评估不符合的数据从数据集移除

@1129571
Copy link
Author

1129571 commented Feb 19, 2024

将offset和confidence评估不符合的数据从数据集移除?

是的,将offset和confidence评估不符合的数据从数据集移除

感谢,我现在就去试试

@1129571
Copy link
Author

1129571 commented Feb 19, 2024

将offset和confidence评估不符合的数据从数据集移除?

是的,将offset和confidence评估不符合的数据从数据集移除

兄弟,我有个新的疑惑想向你确认一下,offset据作者说[-1,1]是最合适的,那么confidence呢。应该怎么判断是否符合,大于3?
#93

@tailangjun
Copy link

将offset和confidence评估不符合的数据从数据集移除?

是的,将offset和confidence评估不符合的数据从数据集移除

兄弟,我有个新的疑惑想向你确认一下,offset据作者说[-1,1]是最合适的,那么confidence呢。应该怎么判断是否符合,大于3? #93

作者说的是 [6,9],我用的是 >= 5

@1129571
Copy link
Author

1129571 commented Feb 19, 2024

将offset和confidence评估不符合的数据从数据集移除?

是的,将offset和confidence评估不符合的数据从数据集移除

兄弟,我有个新的疑惑想向你确认一下,offset据作者说[-1,1]是最合适的,那么confidence呢。应该怎么判断是否符合,大于3? #93

作者说的是 [6,9],我用的是 >= 5

万分感谢

@1129571
Copy link
Author

1129571 commented Feb 21, 2024

将offset和confidence评估不符合的数据从数据集移除?

是的,将offset和confidence评估不符合的数据从数据集移除

兄弟,我有个新的疑惑想向你确认一下,offset据作者说[-1,1]是最合适的,那么confidence呢。应该怎么判断是否符合,大于3? #93

作者说的是 [6,9],我用的是 >= 5

太郎兄你好,我已经参考evaluation/scores_LSE/calculate_scores_LRS.py,将不满足offset[-1, 1], confidence[6, 9]得数据进行了移除,几乎去掉了lrs2中1/3的数据,但是我的鉴别器损失还是卡在0.69附近,你对此有什么经验或建议吗
ce05884dd8e3a644643aef17dc7e72a

@tailangjun
Copy link

  1. 是不是每句话进行一切割,英文控制在3~5秒
  2. 视频是否25帧
  3. 音频是否16000赫兹采样
  4. 检测音视频是否对齐

@prometheus-alien
Copy link

  1. 是不是每句话进行一切割,英文控制在3~5秒
  2. 视频是否25帧
  3. 音频是否16000赫兹采样
  4. 检测音视频是否对齐

你好,请问是否需要训练中文的syncnet模型?

@jtTATannn
Copy link

将offset和confidence评估不符合的数据从数据集移除?

是的,将offset和confidence评估不符合的数据从数据集移除

兄弟,我有个新的疑惑想向你确认一下,offset据作者说[-1,1]是最合适的,那么confidence呢。应该怎么判断是否符合,大于3? #93

作者说的是 [6,9],我用的是 >= 5

太郎兄你好,我已经参考evaluation/scores_LSE/calculate_scores_LRS.py,将不满足offset[-1, 1], confidence[6, 9]得数据进行了移除,几乎去掉了lrs2中1/3的数据,但是我的鉴别器损失还是卡在0.69附近,你对此有什么经验或建议吗 ce05884dd8e3a644643aef17dc7e72a

问题解决了吗,我也遇到同样的问题了,请教一下解决方案

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants