XAI Image Detector - Explainable AI Analysis

Upload Image for Analysis

Leverage Explainable AI to understand the reasoning behind AI-generated image detection.

upload_file

Drag & drop your image(s) here

Click to browse

JPG, JPEG, PNG, WEBP accepted (Max 10MB per file)

Analysis Results

Uploaded Images Summary

[Upload images to see previews here]

Overall Summary

[Summary statistics will appear here after analysis]

{/* Added id and hidden class */}

Selected Image & Saliency

{/* Changed title */}

Image and heatmap will appear here after upload

Heatmap Legend

Indicates image regions influencing the model's prediction.

Blue

Red

Lower Influence Higher Influence

[Typical representation, colors depend on XAI method]

Detection Summary

Predicted Label: N/A

Confidence Score: N/A

Model Used: N/A

Explanation Methods

Saliency Map (e.g., Grad-CAM)

Visualizes important image regions for the prediction, shown as the heatmap overlay on the left.

[Saliency map overlay shown on image]

LIME Explanation

Highlights image segments (superpixels) contributing most to the prediction for this instance.

[LIME visualization placeholder]

SHAP Values

Provides a unified measure of feature importance, showing the contribution of each feature (pixel/region) compared to a baseline.

[SHAP visualization placeholder]

Counterfactual Explanations

Shows minimally modified versions of the image that would change the prediction (e.g., "What change makes it predict 'Real'?").

[Counterfactual examples placeholder]

Explanation Summary

A natural language summary generated from the XAI methods will appear here, explaining the key factors behind the model's decision for this specific image.

[Summary currently unavailable...]

Technical Details & Performance

Layer-wise Relevance Propagation (LRP)

LRP decomposes the prediction decision backward through the network layers to assign relevance scores to input features.

[LRP visualization placeholder]

Attention Mechanism Visualizations

If the model uses attention layers (e.g., in Transformers), this visualizes where the model "focused" when processing the image.

[Attention map placeholder]

Metadata Analysis Results

Analysis of image metadata (e.g., EXIF) can sometimes reveal clues about image origin or manipulation, if available.

No metadata available or analyzed.

Model Architecture

Overview of the neural network architecture used for detection (e.g., CNN, Vision Transformer).

[Model architecture diagram/description placeholder]

API Documentation (Example)

Example of how to use the XAI Detector API programmatically.

POST /analyze
Headers:
  Content-Type: application/json
  Authorization: Bearer YOUR_API_KEY

Body:
{
  "image": "base64_encoded_string_or_url"
}

Response (200 OK):
{
  "prediction": "AI-Generated",
  "confidence_score": 0.975,
  "model_used": "EfficientNet-B7 + XAI",
  "explanations": {
    "saliency_map": "data:image/png;base64,...",
    "lime_regions": [...],
    "shap_values": [...],
    "text_summary": "...",
    "counterfactuals": [...]
  },
  "metrics": { ... }
}

[Note: This is illustrative example documentation.]

Overall Model Performance

Metric	Value
Accuracy	N/A
Precision	N/A
Recall	N/A
F1-Score	N/A

Metrics based on evaluation dataset.

Comparative Analysis

Compare the subtle differences between a real image and an AI-generated counterpart using the slider. (Placeholder images used).

Adjust Comparison View

Real Image AI Generated

Slide the handle to reveal more of the AI-generated image overlay.