README.md

---
license: mit
base_model:
- timm/swin_base_patch4_window7_224.ms_in22k_ft_in1k
pipeline_tag: image-classification
library_name: timm
---

# PowerPoint slide classifier

This is a classifier to classify 5 types of PowerPoint slide layouts. Finetuned from `timm/swin_base_patch4_window7_224.ms_in22k_ft_in1k` and trained on 10k powerpoint slide images.

* `0`: Common content slide
* `1`: End slide
* `2`: Start slide
* `3`: Subtitle slide
* `4`: Subtitle list slide

## Usage

### Install timm and dependencies

```bash
pip install timm==1.0.15 torch==2.7.0 torchvision==0.22.0
```

### Inference

Use the following code to classify images from a folder.

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

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
image_folder = 'path_to_images'

transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225]
    )
])

model = timm.create_model('swin_base_patch4_window7_224', pretrained=False, num_classes=5)
model.load_state_dict(torch.load('pytorch_model.bin'))
model.to(device)
model.eval()

image_files = [f for f in os.listdir(image_folder) if f.lower().endswith('.png')]

idx_to_class = {
    0: 'content',
    1: 'end',
    2: 'start',
    3: 'subt',
    4: 'subtl'
}

with torch.no_grad():
    for image_name in image_files:
        image_path = os.path.join(image_folder, image_name)
        image = Image.open(image_path).convert('RGB')
        input_tensor = transform(image).unsqueeze(0).to(device)

        output = model(input_tensor)
        predicted_class = torch.argmax(output, dim=1).item()
        predicted_label = idx_to_class[predicted_class]

        print(f"{image_name} --> {predicted_label}")
```