README.md

---
base_model:
  - google/vit-base-patch16-224-in21k
library_name: transformers
tags:
  - image-classification
  - vision-transformer
  - just-for-fun
---

# MaxVision: Max vs. Not Max Classifier

## Model Overview

**MaxVision** is a fun, hobby AI vision classifier designed to distinguish between images of Max, a black and white
sprocker spaniel, and all other images. The model has been trained using personal photos of Max and general images of
other dogs and non-dog subjects to improve its classification accuracy. It is intended purely for personal and
experimental use.

## Model Details

- **Developed by:** Patrick Skillen
- **Use Case:** Identifying whether an image contains Max
- **Architecture:** Based on a fine-tuned vision transformer (ViT)
- **Training Dataset:** Curated personal dataset of Max and various non-Max images
- **Framework:** PyTorch with Hugging Face Transformers
- **Training Platform:** Google Colab
- **Labels:**
    - `0`: Max
    - `1`: Not Max

## Intended Use

This model is built as a fun, personal experiment in AI/ML and image classification. It is not intended for commercial
applications, biometric identification, or general dog breed classification.

## Limitations & Biases

- The model is heavily biased toward distinguishing Max from non-Max images and is not robust for identifying specific
  breeds or other dogs.
- Performance may degrade on images with low resolution, extreme lighting conditions, or unusual poses.
- Limited dataset size and personal image selection may affect generalizability.

## How to Use

Try it in the HF Space at https://huggingface.co/spaces/paddeh/is-it-max

To use the model, you can run inference using the Hugging Face `transformers` or `timm` library, depending on the model
backbone. Below is a sample inference script:

```python
from transformers import pipeline

classifier = pipeline("image-classification", model="paddeh/is-it-max")

result = classifier("path/to/image.jpg")
print("Max" if prediction.item() == 0 else "Not Max")
```

Alternatively, with `torchvision`:

```python
import torch
from torchvision import transforms
from transformers import ViTForImageClassification, ViTImageProcessor
from PIL import Image

model = ViTForImageClassification.from_pretrained('model.safetensors')
model.eval()
processor = ViTImageProcessor.from_pretrained(model_path)

transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=feature_extractor.image_mean, std=feature_extractor.image_std),
])

image = Image.open("path/to/image.jpg")
image = transform(image).unsqueeze(0)

with torch.no_grad():
    output = model(image)

prediction = torch.argmax(output, dim=1)
print("Max" if prediction.item() == 0 else "Not Max")
```

## Model Performance

As this is a personal hobby project, there is no formal benchmark, but the model has been tested informally using
validation images from Max’s personal collection and various other dog breeds.

## Ethical Considerations

Since this model is built for personal use, there are no significant ethical concerns. However, users should be mindful
of data privacy and not use the model for unauthorized biometric identification of pets or people.

## Future Improvements

- Expand the dataset with more diverse images of Max in different lighting conditions and settings.
- Improve augmentation techniques to enhance robustness.
- Fine-tune using more advanced architectures like CLIP or Swin Transformer for better accuracy.

---

**Disclaimer:** This model is intended for personal and educational use only. It is not designed for commercial
applications or general-purpose image recognition.