|
|
--- |
|
|
license: apache-2.0 |
|
|
language: |
|
|
- en |
|
|
base_model: |
|
|
- google/siglip2-base-patch16-224 |
|
|
pipeline_tag: image-classification |
|
|
library_name: transformers |
|
|
datasets: |
|
|
- KarteeMonkey/Demo |
|
|
tags: |
|
|
- Game |
|
|
- Moderation |
|
|
- art |
|
|
- code |
|
|
--- |
|
|
|
|
|
 |
|
|
|
|
|
# appy-mod-beta1 |
|
|
|
|
|
> **`appy-mod-beta1`** is a **vision-language encoder model** fine-tuned from `siglip2-base-patch16-224` for **binary image classification**. The model is trained to perform **game content moderation**, specifically classifying visual content as either **safe (good)** or **unsafe (bad)**. It utilizes the `SiglipForImageClassification` architecture. |
|
|
|
|
|
> \[!note] |
|
|
> SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features |
|
|
> [https://arxiv.org/pdf/2502.14786](https://arxiv.org/pdf/2502.14786) |
|
|
|
|
|
```py |
|
|
Classification Report: |
|
|
precision recall f1-score support |
|
|
|
|
|
bad 0.9763 0.9140 0.9441 1755 |
|
|
good 0.9279 0.9803 0.9534 1983 |
|
|
|
|
|
accuracy 0.9492 3738 |
|
|
macro avg 0.9521 0.9471 0.9487 3738 |
|
|
weighted avg 0.9506 0.9492 0.9490 3738 |
|
|
``` |
|
|
|
|
|
 |
|
|
|
|
|
--- |
|
|
|
|
|
## Label Space: 2 Classes |
|
|
|
|
|
``` |
|
|
Class 0: bad (Unsafe content) |
|
|
Class 1: good (Safe content) |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## Install Dependencies |
|
|
|
|
|
```bash |
|
|
pip install transformers torch pillow gradio hf_xet |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## Inference Code |
|
|
|
|
|
```python |
|
|
import gradio as gr |
|
|
from transformers import AutoImageProcessor, SiglipForImageClassification |
|
|
from PIL import Image |
|
|
import torch |
|
|
|
|
|
# Load model and processor |
|
|
model_name = "KarteeMonkey/appy-mod-beta1" # Update this if using a different path |
|
|
model = SiglipForImageClassification.from_pretrained(model_name) |
|
|
processor = AutoImageProcessor.from_pretrained(model_name) |
|
|
|
|
|
# Label mapping |
|
|
id2label = { |
|
|
"0": "bad", |
|
|
"1": "good" |
|
|
} |
|
|
|
|
|
def classify_watermark(image): |
|
|
image = Image.fromarray(image).convert("RGB") |
|
|
inputs = processor(images=image, return_tensors="pt") |
|
|
|
|
|
with torch.no_grad(): |
|
|
outputs = model(**inputs) |
|
|
logits = outputs.logits |
|
|
probs = torch.nn.functional.softmax(logits, dim=1).squeeze().tolist() |
|
|
|
|
|
prediction = { |
|
|
id2label[str(i)]: round(probs[i], 3) for i in range(len(probs)) |
|
|
} |
|
|
|
|
|
return prediction |
|
|
|
|
|
# Gradio Interface |
|
|
iface = gr.Interface( |
|
|
fn=classify_watermark, |
|
|
inputs=gr.Image(type="numpy"), |
|
|
outputs=gr.Label(num_top_classes=2, label="Game Anomaly Detection"), |
|
|
title="Game Anomaly Detection SigLIP2", |
|
|
description="Upload an image to detect whether it contains a anomaly." |
|
|
) |
|
|
|
|
|
if __name__ == "__main__": |
|
|
iface.launch() |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## Intended Use |
|
|
|
|
|
`appy-mod-beta1` is designed for: |
|
|
|
|
|
* **Game Content Moderation** – Automated moderation of user-generated or in-game visual content. |
|
|
* **Parental Control Tools** – Supports identifying unsafe or inappropriate content in children’s games. |
|
|
* **Online Game Platforms** – Enables scalable and automatic screening of images uploaded by users. |
|
|
* **Community Safety** – Helps maintain safe and compliant visual environments in multiplayer games and forums. |
|
|
* **AI Moderation Research** – A sample project for applying vision-language models to safety-critical applications. |
