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

Run OpenSeeFaceLandmarks #70

Open
namhai1810 opened this issue Jun 25, 2024 · 8 comments
Open

Run OpenSeeFaceLandmarks #70

namhai1810 opened this issue Jun 25, 2024 · 8 comments

Comments

@namhai1810
Copy link

Great work bro. However, why I run file model.py got this issued that cannot mapping **kwargs
image
image

image

@emilianavt
Copy link
Owner

Make sure that you are on commit c450c12ae6ffb1757f62dde3c2765da3c10f6def of geffnet.

@namhai1810
Copy link
Author

namhai1810 commented Jun 25, 2024

oh could you give me version of your geffnet, I only use pip install geffnet(its version 1.0.2)

@emilianavt
Copy link
Owner

I'd recommend cloning it from github and installing it that way.

@namhai1810
Copy link
Author

I have solve this problem, the main reason is the the latest geffnet. Many thanks

@namhai1810
Copy link
Author

namhai1810 commented Jun 27, 2024

I have observed the output of OpenSeeFaceLandmarks during training, which is of the shape (1, 198, 28, 28), with an input shape of (1, 3, 224, 224). However, our labels consist of 66 landmarks. What does the dimension 198 represent in the output? Additionally, could you provide some guidance on how to create DataLoaders for our data and process labels, given that our labels are in the form of (x, y) coordinates of landmarks?

@emilianavt
Copy link
Owner

I recommend reading through #1. Some discussion of such topics has taken place there. Regarding your question for the shapes, referring to the landmarks function posted there may help.

@namhai1810
Copy link
Author

I have created a heatmap of coordinates (x, y) based on a Gaussian distribution. However, I am encountering difficulties in converting these (x, y) coordinates into X_offset_map and Y_offset_map. Could you please provide guidance on how to create these offset maps?

Furthermore, I reviewed your landmark code and noticed a function named logit_arr to reverse the effect of the sigmoid function. Does the scale factor affect the way we create the offset map? @emilianavt

@emilianavt
Copy link
Owner

For the offset maps, you can find a nice explanation of how they work here in the "A Technical Deep Dive" section. The specific resolution numbers are different, but the technique is the same. The sigmoid function and scale factor adjust the steepness of the gradient on the offset masks.

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