The weights for StableNormal-turbo

[BIT-DYN]

Such an amazing job!

I have downloaded the original weights for StableNormal followed #14.

But where can I download the weights file for StableNormal-turbo?
I want to test which of the two is better for our scenes.
Thanks!

[hugoycj]

We are currently conducting beta testing for StableNormal_turbo to optimize the balance between sharpness and accuracy. You can try the demo on Hugging Face.
For local testing, please use the following commands:

import torch
from PIL import Image

# Load an image
input_image = Image.open("path/to/your/image.jpg")

# Create predictor instance
predictor = torch.hub.load("hugoycj/StableNormal", "StableNormal_turbo", trust_repo=True, yoso_version='yoso-normal-v1-5')

# Generate normal map using alpha channel for masking
normal_map = predictor(rgba_image, data_type="object")  # Will mask out background, if alpha channel is avalible, else use birefnet
normal_map = predictor(rgba_image, data_type="outdoor")  # Will use Mask2Former to mask out sky and plants
normal_map = predictor(rgba_image, data_type="indoor") # Will not mask out

# Apply the model to the image
normal_image = predictor(input_image)

# Save or display the result
normal_image.save("output/normal_map.png")

[BIT-DYN]

Thank you for your quick reply!

Since my terminal doesn't have direct access to huggingface (but the browser does).
So I chose to download the v1-5 weights locally, but there seems to be some problems.

v1-0 weights is available, but the results are very unsatisfactory.

Do you have any suggestions?

[xingchen2022]

Thank you for your quick reply!

Since my terminal doesn't have direct access to huggingface (but the browser does). So I chose to download the v1-5 weights locally, but there seems to be some problems.

v1-0 weights is available, but the results are very unsatisfactory.

Do you have any suggestions?

same question, have you solved this？

[xingchen2022]

@BIT-DYN hi, have you find the proper way to reproduce the result like online demo？I have tried different weights in #31

[BIT-DYN]

@BIT-DYN hi, have you find the proper way to reproduce the result like online demo？I have tried different weights in #31

Unfortunately, I have not solved this problem so far.

[hugoycj]

@BIT-DYN Sorry for the late reply. Could you share your input image and the normal result from the HuggingFace demo? StableNormal may fail for defocused, horizontally oriented or cartoon-style images. It might be our model's problem. Thanks.