search

IDEA-Research / HumanSD

[ICCV 2023] The official implementation of paper "HumanSD: A Native Skeleton-Guided Diffusion Model for Human Image Generation"
Apache License 2.0
271 stars 18 forks source link

About the data quality #21

Closed leesky1c closed 11 months ago

leesky1c commented 11 months ago

Hi Xuan,

I used the utils/download_data.py script from the repository to download data via the Laion-Human's URL.

After downloading 1000 images, I randomly sampled several images which looks like:

000002238 000001138 000001038

I would like to confirm if I missed any steps or where the problem occurred.

Thanks.

juxuan27 commented 11 months ago

Hi, @leeskyed . Thank you for your focus! The images are selected with pose estimation score>0 thus containing some images without human. This is to ensure the flexibility of using different pose estimation score in training. You can try using the given score>0.02 in our code as a selection threshold, you will find much less false detection.

leesky1c commented 11 months ago

Thanks, I will give it a try! :)