texture-brush / README.md
Druudik1's picture
Update README.md
6294185 verified
|
Raw History Blame Contribute Delete
1.53 kB
---
license: mit
---
This repository contains trained `GeneralNeuralRenderer` that learned to mimic `TextureBrush` (these are
components of my [Stylized Neural Painting paper reimplementation](https://github.com/Druudik/stylized-neural-painting)).
Use the following code snippet to load it and create the corresponding `TextureBrush`:
```python
from PIL import Image
from huggingface_hub import hf_hub_download
from painting.brushes import TextureBrush, GeneralNeuralRenderer
import torchvision.transforms.functional as TF
def load_texture(source: str, device: str):
img = Image.open(source).convert("L")
tensor = TF.pil_to_tensor(img).float() / 255.0
return tensor.to(device)
# Specify your device and output canvas size.
device = "cuda"
out_canvas_size = 512
brush = TextureBrush(
large_vertical=load_texture("./assets/textures/large_vertical_brush.png", device),
large_horizontal=load_texture("./assets/textures/large_horizontal_brush.png", device),
small_vertical=load_texture("./assets/textures/small_vertical_brush.png", device),
small_horizontal=load_texture("./assets/textures/small_horizontal_brush.png", device),
canvas_size=out_canvas_size,
device=device,
)
differentiable_brush = GeneralNeuralRenderer(
brush_params_count=brush.brush_params_count,
canvas_size=128,
device=device,
)
weights = hf_hub_download(repo_id="Druudik1/texture-brush", filename="neural_renderer.pt")
differentiable_brush.load_state_dict(
torch.load(weights, map_location=device, weights_only=True)
)
```