README.md

---
license: cc-by-nc-4.0
language:
- en
base_model:
- facebook/metaclip-2-worldwide-s16
pipeline_tag: image-classification
library_name: transformers
tags:
- text-generation-inference
- Content-filtering
model-index:
- name: Nsfw_Image_Detection_OSS
  results:
  - task:
      type: image-classification
    dataset:
      type: evaluation
      name: NSFW Image Detection Benchmark
    metrics:
    - type: accuracy
      value: 0.8918
      name: Accuracy

    - type: f1
      value: 0.9071
      name: F1 (NSFW)

    - type: precision
      value: 0.9047
      name: Precision (NSFW)

    - type: recall
      value: 0.9094
      name: Recall (NSFW)

    - type: f1
      value: 0.8705
      name: F1 (SFW)

    - type: precision
      value: 0.8736
      name: Precision (SFW)

    - type: recall
      value: 0.8673
      name: Recall (SFW)

    - type: f1_macro
      value: 0.8888
      name: Macro F1

    - type: f1_weighted
      value: 0.8917
      name: Weighted F1
---

![1](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/0d0tOhRfsB9KEn4iuU8Zn.png)

# **Nsfw_Image_Detection_OSS**

> **Nsfw_Image_Detection_OSS** is an image classification vision-language encoder model fine-tuned from **[facebook/metaclip-2-worldwide-s16](https://huggingface.co/facebook/metaclip-2-worldwide-s16)** for a **binary NSFW detection task**.
> It is designed to classify whether an image is **Safe For Work (SFW)** or **Not Safe For Work (NSFW)** using the **MetaClip2ForImageClassification** architecture.

> [!note]
> **MetaCLIP 2: A Worldwide Scaling Recipe**
> [https://huggingface.co/papers/2507.22062](https://huggingface.co/papers/2507.22062)

## Evaluation Report (Self-Reported)

```py
Classification report:

              precision    recall  f1-score   support

         SFW     0.8736    0.8673    0.8705     11103
        NSFW     0.9047    0.9094    0.9071     15380

    accuracy                         0.8918     26483
   macro avg     0.8892    0.8884    0.8888     26483
weighted avg     0.8917    0.8918    0.8917     26483
```

![download](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/T1Hp_Yuhew7gUJJH-Prmy.png)

# **Label Mapping**

The model categorizes images into two classes:

* **Class 0:** **SFW**
* **Class 1:** **NSFW**

```json
{
  "id2label": {
    "0": "SFW",
    "1": "NSFW"
  },
  "label2id": {
    "SFW": 0,
    "NSFW": 1
  }
}
```

# **Run with Transformers**

```python
!pip install -q transformers torch pillow gradio
```

```python
import gradio as gr
import torch
from transformers import AutoImageProcessor, AutoModelForImageClassification
from PIL import Image

# Model name from Hugging Face Hub
model_name = "prithivMLmods/Nsfw_Image_Detection_OSS"

# Load processor and model
processor = AutoImageProcessor.from_pretrained(model_name)
model = AutoModelForImageClassification.from_pretrained(model_name)
model.eval()

# Define labels
LABELS = {
    0: "SFW",
    1: "NSFW"
}

def nsfw_detection(image):
    """Predict whether an image is SFW or NSFW."""
    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()

    predictions = {LABELS[i]: round(probs[i], 3) for i in range(len(probs))}
    return predictions

# Build Gradio interface
iface = gr.Interface(
    fn=nsfw_detection,
    inputs=gr.Image(type="numpy", label="Upload Image"),
    outputs=gr.Label(label="NSFW Detection Probabilities"),
    title="NSFW Image Detection (MetaCLIP-2)",
    description="Upload an image to classify whether it is Safe For Work (SFW) or Not Safe For Work (NSFW)."
)

# Launch app
if __name__ == "__main__":
    iface.launch()
```

# **Intended Use**

The **Nsfw_Image_Detection_OSS** model is designed to classify images into **SFW or NSFW categories**.

Potential use cases include:

* **Content Moderation:** Automated filtering of unsafe or adult content.
* **Social Media Platforms:** Preventing the upload of explicit media.
* **Enterprise Safety:** Ensuring workplace-appropriate content in shared environments.
* **Dataset Filtering:** Cleaning large-scale image datasets before training.
* **Parental Control Systems:** Blocking inappropriate visual material.