Submissions will be evaluated based on the following criteria:

1. Accuracy and Robustness (65%)
Real vs. AI Classification accuracy as per the leaderboard
2. Inference Speed (15%)
Model execution time on the standardized test hardware
Efficiency of your implementation
3. Explainability and Presentation (20%)
Participants must incorporate at least one explainability method:
Attention visualization (e.g., Grad-CAM)
Natural language explanations/captions for classifications
Feature importance visualization
Decision boundary analysis

Your final submission must be a CSV file with the following format:

```filename,class```
```image_001.png,0```
```image_002.png,1```
Note: 0 corresponds to the label ai and 1 corresponds to real images
...
A sample file (sample.csv) has been provided for reference.