kennyvoo / face-occlusion-generation Goto Github PK

Hi!
Your work is awesome! I tried your provided model and the performance is excellent.
However the inference time is a little bit long. Could you release some light pretrained model(unet or pspnet) which could be
deployed on CPU-backend devices?

Could you please provide code to apply this to datasets other than CelebAHQ.
For example, I would like to apply this to VGGface2 to make a natural occluded VGGFace2 dataset

Thank you!

#14 (comment)

Hi,
Is there a License Restrictions on the dataset? Could it be used for commercial purposes?

Hello,

Do you have code to create a mask from the occluded face?

Thanks!

Hi,
Thanks for your repo. I', wondering do we needs batch here?
for batch in range (batch_size):
I guess you just put one src image and one occ image as a pair into
color_transfer_sot

BR,
Ziv

Hello,

Great work and congratulations on getting accepted to CVPRW.

I wanted to request you the trained model checkpoints of DeepLabV3, etc. If you could share that would be amazing as I am trying to get one my project working and your work got me really excited.

Thanks.

Hi Dr.
I am writing to ask if you can release your code ASAP to compare with.

looking for faceset with all images occluded with hands and random stuff and everything

I am wondering if there is a complete code（train.py）to train with the datasets CelebAMask-HQ and CelebAMask-HQ-masks_corrected you provided?

Thank you

Hello, sir. Congratulations on your success. I have some questions to ask. It refers to whether you can change some codes to generate occlusion without changing the original image. Because some pictures will lighten the color of the un occluded part after being occluded. Thank you for your answer

hi，
I use the natocc script to generate hand occlusion pictures, but the generated pictures only have 2 categories (face and background). Where is the code for changing the category? I want to superimpose only the hand occlusion, and the rest remain the previous categories

Hi thanks for the great work about occlusion-aware augmentation method.

is there a plan for sharing training code or a pretrained weight of models (segformer) in the paper?

Hi,
could you please provide a pretrained model of segformer or any other tested model?

thank you

Hi, is it okay to PR this dataset to mmsegmentation?

I have been using this dataset since you publish the paper and it is very helpful to my projects.

I think it would be nice if mmsegmentation supports this dataset