---
title: Fumo Classifier
emoji: 🧸
colorFrom: blue
colorTo: pink
sdk: gradio
sdk_version: 5.9.1
app_file: app.py
pinned: false
---

# Fumo Image Classifier

## Overview
A LoRA for `google/siglip-so400m-patch14-384` trained to identify Fumo dolls (Gift) in images. This model achieves 99.28% accuracy in fumo detection with a bias towards false negatives.

## Model Details
- Base Model: `google/siglip-so400m-patch14-384`
- Training Method: LoRA fine-tuning (r=16, alpha=32)
- Target Modules: Attention (q_proj, v_proj) and FFN (fc1, fc2)
- Input: Images (384x384 pixels)
- Output: Binary classification (fumo/not fumo)
- Accuracy: 99.28%

## Training Specifications
- Training Device: NVIDIA RTX 3050 Ti
- Training Time: ~2 hours
- Epochs: 15
- Batch Size: 10
- Optimizer: AdamW (lr=1.3e-5, weight_decay=0.001)
- Data Augmentation:
* Random vertical flip (p=0.5)
* Color jitter (hue=0.5)

## Usage
```python
from transformers import AutoProcessor, AutoModel
import torch
from peft import PeftModel, PeftConfig

# Load the base model and processor
processor = AutoProcessor.from_pretrained("google/siglip-so400m-patch14-384")
model = AutoModel.from_pretrained("google/siglip-so400m-patch14-384")

# Load LoRA weights
model = PeftModel.from_pretrained(model, "houraisan/fumo-classifier")

def classify_image(image_path):
    # Preprocess image
    inputs = processor(images=image_path, return_tensors="pt")

    # Get prediction
    with torch.no_grad():
        outputs = model(**inputs)

    # Process outputs
    # Classification logic here
```

## Model Architecture
```python
LoraConfig(
    r=16,
    lora_alpha=32,
    target_modules=["q_proj", "v_proj", "fc1", "fc2"],
    lora_dropout=0.1,
    bias="none",
)
```

## Citations
```bibtex
@misc{houraisan2024fumo,
  author = {Houraisan, Kaguya},
  title = {Fumo Image Classifier},
  year = {2024},
  publisher = {Hugging Face Spaces},
  journal = {Imperial Lunar Institute Research Repository}
}
```

## License
MIT License

Copyright (c) 2024 Kaguya Houraisan

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

## Acknowledgments
- Google for the base SigLIP model
- Hugging Face for hosting infrastructure