README.md

---
license: mit
datasets:
  - zobeir/GoldNet
tags:
  - image-classification
  - pytorch
  - vision-transformer
  - counterfeit-detection
  - gold
  - fine-grained-recognition
language:
  - en
---

# GoldNet Model Weights

Trained checkpoints for **GoldFormer** and baseline models from the paper:

> **GoldFormer: A Texture-Aware Vision Transformer-based Algorithm for Detecting Near-Identical Images**  
> Z. Raisi, *Algorithms* (MDPI), under review.  
> Code & dataset: [github.com/zobeirraisi/GoldNet](https://github.com/zobeirraisi/GoldNet)

## Task

Binary image classification — **authentic vs. counterfeit gold items** — from ordinary smartphone photographs. The two classes are near-identical to the eye; trained experts reached 89.80% accuracy on a blind subset.

## Available Checkpoints (`weights/`)

| File | Model | Accuracy (5-fold CV) |
|---|---|---|
| `GoldFormer_best.pth` | GoldFormer (CNN + Swin-T + TAAG) | 94.69 ± 0.79% |
| `Swin_T_best.pth` | Swin Transformer-Tiny | 94.31 ± 0.78% |
| `ViT_B16_best.pth` | ViT-B/16 | 94.31 ± 0.94% |
| `ResNet101_best.pth` | ResNet-101 | 92.29 ± 1.01% |
| `ResNet50_best.pth` | ResNet-50 | — |
| `ResNet18_best.pth` | ResNet-18 | — |
| `DenseNet121_best.pth` | DenseNet-121 | — |
| `EfficientNet_B3_best.pth` | EfficientNet-B3 | — |
| `EfficientNet_B0_best.pth` | EfficientNet-B0 | — |
| `MobileNet_V2_best.pth` | MobileNet-V2 | — |

All models trained with 5-fold stratified cross-validation, AdamW, AMP (bfloat16), freeze-then-unfreeze fine-tuning on the GoldNet dataset (2,127 images, 1,044 authentic / 1,083 counterfeit).

## Usage

```python
import torch
from torchvision import transforms
from PIL import Image

# Download weights
# bash fetch_weights.sh   (from the GitHub repo)

# Load a checkpoint
model = torch.load("weights/GoldFormer_best.pth", weights_only=True)
model.eval()

transform = transforms.Compose([
    transforms.Resize((299, 299)),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406],
                         [0.229, 0.224, 0.225]),
])

img = Image.open("your_image.jpg").convert("RGB")
x = transform(img).unsqueeze(0)

with torch.no_grad():
    logits = model(x)
    prob_authentic = torch.softmax(logits, dim=1)[0, 0].item()
    print(f"P(authentic) = {prob_authentic:.3f}")
```

> **Note:** All baseline models use 224×224 input. GoldFormer uses 299×299.  
> The `models.py` class definitions are in the [GitHub repo](https://github.com/zobeirraisi/GoldNet).

## Citation

```bibtex
@article{raisi2026goldformer,
  title   = {GoldFormer: A Texture-Aware Vision Transformer-based Algorithm
             for Detecting Near-Identical Images},
  author  = {Raisi, Zobeir},
  journal = {Algorithms},
  year    = {2026},
  note    = {Under review}
}
```

## License

Model weights: [MIT License](https://github.com/zobeirraisi/GoldNet/blob/main/LICENSE)  
Dataset: [CC BY 4.0](https://github.com/zobeirraisi/GoldNet/blob/main/LICENSE-DATA)