Skip to content
This repository has been archived by the owner on Jan 10, 2023. It is now read-only.

kp2uv model #5

Open
BadourAlBahar opened this issue Sep 6, 2021 · 4 comments
Open

kp2uv model #5

BadourAlBahar opened this issue Sep 6, 2021 · 4 comments

Comments

@BadourAlBahar
Copy link

Hi Erika,

Congratulations on the great work. Very impressive.

I am trying to use the kp2uv model but I get weird results. Please find attached the kp_im and uv_output.

Any ideas why I am getting this.

Thanks!

kp_im
output_uv

@erikalu
Copy link
Contributor

erikalu commented Sep 8, 2021

Hi, this result looks reasonable to me. The shape of the head is a bit strange, but our layered neural rendering model can correct for slight issues like that, so this output is sufficient for our use case. Is that what you meant by the result looks 'weird'?

@BadourAlBahar
Copy link
Author

Hi Erika,

The “weird” part for me is the pink color. I thought it should look blue/green.

I was also wondering how you trained this model. What was the loss of the model?

@erikalu
Copy link
Contributor

erikalu commented Sep 9, 2021

Oh the pink is just the body part ID. The RGB channels correspond to IUV, where I is an integer 1-25 ( *10 for visualization purposes).

The losses are the same as in the original DensePose paper (see section 3.1).

@BadourAlBahar
Copy link
Author

Thank you for the information Erika. Can you share the code for training the kp2dp model?

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

No branches or pull requests

2 participants