---
license: mit
pipeline_tag: image-to-image
tags:
- style-transfer
- pytorch
---

# Fast Neural Style Transfer — Starry Night

This repository contains weights for a Fast Neural Style Transfer network based on Johnson et al. It is trained on the COCO val2017 dataset to instantly apply Vincent van Gogh's *The Starry Night* style to any input image.

## Style Transfer Preview


| Content Image | Stylized Output |
| :---: | :---: |
| <img src="before.jpg" width="400"> | <img src="after.jpg" width="400"> |

---

## How to Use Programmatically

You can run inference using the official `huggingface_hub` utility library. The script automatically downloads your weights file directly from the cloud and applies the necessary ImageNet normalization matching the training routine.

### Dependencies
Ensure you have the required packages installed:
```bash
pip install torch torchvision pillow huggingface_hub
```

### Inference Script (`inference.py`)
Save the following code as `inference.py`. You can run it via terminal with `python inference.py your_image.jpg`.

```python
import sys
import torch
import torch.nn as nn
from PIL import Image
from torchvision import transforms
from torchvision.utils import save_image
from huggingface_hub import hf_hub_download

# ── CONFIG ───────────────────────────────────────────────────
REPO_ID  = "Rohanify/Brawnz-StyleTransferSN"
FILENAME = "pytorch_model.bin"
IMG_SIZE = 512
DEVICE   = "cuda" if torch.cuda.is_available() else "cpu"
# ─────────────────────────────────────────────────────────────

# ── NATIVE PYTORCH NETWORK DEFINITION ────────────────────────

def conv_bn_relu(in_c, out_c, k, stride=1, pad=0):
    return nn.Sequential(
        nn.ReflectionPad2d(pad),
        nn.Conv2d(in_c, out_c, k, stride),
        nn.InstanceNorm2d(out_c),
        nn.ReLU(inplace=True),
    )

class ResBlock(nn.Module):
    def __init__(self, c):
        super().__init__()
        self.block = nn.Sequential(
            nn.ReflectionPad2d(1),
            nn.Conv2d(c, c, 3),
            nn.InstanceNorm2d(c),
            nn.ReLU(inplace=True),
            nn.ReflectionPad2d(1),
            nn.Conv2d(c, c, 3),
            nn.InstanceNorm2d(c),
        )
    def forward(self, x):
        return x + self.block(x)

class TransformNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.net = nn.Sequential(
            conv_bn_relu(3,   32,  9, pad=4),
            conv_bn_relu(32,  64,  3, stride=2, pad=1),
            conv_bn_relu(64, 128,  3, stride=2, pad=1),
            ResBlock(128), ResBlock(128), ResBlock(128),
            ResBlock(128), ResBlock(128),
            nn.Upsample(scale_factor=2, mode="nearest"),
            conv_bn_relu(128, 64, 3, pad=1),
            nn.Upsample(scale_factor=2, mode="nearest"),
            conv_bn_relu(64, 32, 3, pad=1),
            nn.ReflectionPad2d(4),
            nn.Conv2d(32, 3, 9),
            nn.Tanh(),
        )
    def forward(self, x):
        return self.net(x)

# ── LOAD INPUT IMAGE ─────────────────────────────────────────
if len(sys.argv) < 2:
    print("Usage: python inference.py path_to_input_image.jpg")
    sys.exit(1)

input_path = sys.argv[1]
output_path = "output_styled.jpg"

transform = transforms.Compose([
    transforms.Resize((IMG_SIZE, IMG_SIZE)),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
])

img = Image.open(input_path).convert("RGB")
x   = transform(img).unsqueeze(0).to(DEVICE)

# ── SECURE FILE DOWNLOAD & STATE LOAD ────────────────────────
print("Downloading weights from Hugging Face Hub...")
weights_path = hf_hub_download(repo_id=REPO_ID, filename=FILENAME)

model = TransformNet().to(DEVICE)
model.load_state_dict(torch.load(weights_path, map_location=DEVICE))
model.eval()
print(f"Weights successfully loaded on: {DEVICE}")

# ── RUN INFERENCE ────────────────────────────────────────────
print("Processing style transfer...")
with torch.no_grad():
    out = model(x)

save_image(out[0] * 0.5 + 0.5, output_path)
print(f"Success! Styled image saved to: {output_path}")

```