IndoorOutdoorNet

IndoorOutdoorNet is an image classification vision-language encoder model fine-tuned from google/siglip2-base-patch16-224 for a single-label classification task. It is designed to classify images as either Indoor or Outdoor using the SiglipForImageClassification architecture.

Classification Report:
              precision    recall  f1-score   support

      Indoor     0.9661    0.9554    0.9607      9999
     Outdoor     0.9559    0.9665    0.9612      9999

    accuracy                         0.9609     19998
   macro avg     0.9610    0.9609    0.9609     19998
weighted avg     0.9610    0.9609    0.9609     19998

---

The model categorizes images into 2 environment-related classes:

    Class 0: "Indoor"
    Class 1: "Outdoor"

---

Install dependencies

!pip install -q transformers torch pillow gradio

---

Inference Code

import gradio as gr
from transformers import AutoImageProcessor, SiglipForImageClassification
from PIL import Image
import torch

# Load model and processor
model_name = "prithivMLmods/IndoorOutdoorNet"  # Updated model name
model = SiglipForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)

def classify_environment_image(image):
    """Predicts whether an image is Indoor or Outdoor."""
    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()
    
    labels = {
        "0": "Indoor", "1": "Outdoor"
    }
    predictions = {labels[str(i)]: round(probs[i], 3) for i in range(len(probs))}
    
    return predictions

# Create Gradio interface
iface = gr.Interface(
    fn=classify_environment_image,
    inputs=gr.Image(type="numpy"),
    outputs=gr.Label(label="Prediction Scores"),
    title="IndoorOutdoorNet",
    description="Upload an image to classify it as Indoor or Outdoor."
)

if __name__ == "__main__":
    iface.launch()

---

Intended Use:

The IndoorOutdoorNet model is designed to classify images into indoor or outdoor environments. Potential use cases include:

• Smart Cameras: Detect indoor/outdoor context to adjust settings.
• Dataset Curation: Automatically filter image datasets by setting.
• Robotics & Drones: Environment-aware navigation logic.
• Content Filtering: Moderate or tag environment context in image platforms.