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 train the DAN-Tracking.npz model? #56

Open
zhaofang216 opened this issue Jun 25, 2019 · 3 comments
Open

How to train the DAN-Tracking.npz model? #56

zhaofang216 opened this issue Jun 25, 2019 · 3 comments

Comments

@zhaofang216
Copy link

Hi ,

  Thanks for your codes.

  I have trained a model of 106  landmarks (without tracking) via your codes,  now I want to train the tracking model, but i do not know how to do it.  Can you give me some suggestions?
@MarekKowalski
Copy link
Owner

Hi,

I assume, that by tracking model you mean the one that is able to predict whether it lost tracking or not.
This model was trained by adding a single additional dense layer attached to the penultimate layer of the first stage, you can see it in line 117 of this file.

The layer was trained to distinguish faces from non-faces using the WIDER FACE dataset. Since that dataset contains annotations for all faces in all images, it was easy to generate a large dataset of face and non-face patches. In addition we blurred some of the face patches to simulate motion blur that occurs during tracking.

Does that help?

Marek

@zhaofang216
Copy link
Author

Hi Marek,
Thanks for you kindly reply, and could you give me more details about how to train additional layer? I wonder what the neural network to train the confidence layer is. How many layers the neural network contains?

Thanks.

@MarekKowalski
Copy link
Owner

Hi,
As mentioned in the previous post, the neural network consists of a single layer, you can see it in the file I linked to.

Marek

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

2 participants