Model Card: Military Aircraft Detection with Vision Transformer (ViT)
Model Information
- Model Name: Military Aircraft Image Detection
- Model Type: Vision Transformer (ViT)
Model Overview
- Purpose: The model is designed for the detection and classification of military aircraft in images.
- Intended Use: Military surveillance, object recognition, and security applications.
Model Training
- Training Data: Dataset of military aircraft images collected from Illia56/Military-Aircraft-Detection.
- Data Preprocessing: Random oversampling for class balance, data augmentation (rotation, flip, sharpness adjustment).
- Model Architecture: Vision Transformer (ViT) for image classification.
- Pre-trained Model: google/vit-base-patch16-224-in21k.
Model Evaluation
- Evaluation Metrics:
- Accuracy
- F1 Score
- Confusion Matrix
- Evaluation Dataset: Split from the original dataset for testing.
Class Precision Recall F1-Score Support A10 0.6716 0.7368 0.7027 247 A400M 0.6217 0.6748 0.6472 246 AG600 0.4512 0.9919 0.6203 247 AV8B 0.6618 0.7287 0.6936 247 B1 0.9000 0.6194 0.7338 247 B2 0.7862 0.9231 0.8492 247 B52 0.9528 0.4089 0.5722 247 Be200 0.8333 0.8300 0.8316 247 C130 0.8600 0.1748 0.2905 246 C17 0.5556 0.0405 0.0755 247 C2 0.5845 0.8543 0.6941 247 C5 0.3776 0.7490 0.5020 247 E2 0.8447 0.9028 0.8728 247 E7 0.6000 0.9595 0.7383 247 EF2000 1.0000 0.0364 0.0703 247 F117 0.6005 0.9433 0.7339 247 F14 0.9773 0.1741 0.2955 247 F15 0.2919 0.2186 0.2500 247 F16 0.8333 0.0203 0.0397 246 F18 0.9355 0.2348 0.3754 247 F22 0.4624 0.4980 0.4795 247 F35 0.5373 0.2915 0.3780 247 F4 0.4317 0.2429 0.3109 247 J10 0.8711 0.6842 0.7664 247 J20 0.5049 0.6301 0.5606 246 JAS39 0.4535 0.4737 0.4634 247 KC135 0.8957 0.7683 0.8271 246 MQ9 0.7358 0.8943 0.8073 246 Mig31 0.6080 0.4899 0.5426 247 Mirage2000 0.3245 0.6478 0.4324 247 P3 0.9423 0.3968 0.5584 247 RQ4 0.7166 0.8907 0.7942 247 Rafale 0.3063 0.3968 0.3457 247 SR71 0.7824 0.7571 0.7695 247 Su25 1.0000 0.3618 0.5313 246 Su34 0.5340 0.8583 0.6584 247 Su57 0.6143 0.7317 0.6679 246 Tornado 0.6883 0.2146 0.3272 247 Tu160 0.8000 0.8421 0.8205 247 Tu95 0.8340 0.8543 0.8440 247 U2 0.9371 0.6032 0.7340 247 US2 0.7074 0.6559 0.6807 247 V22 0.7212 0.9109 0.8050 247 Vulcan 0.3343 0.8947 0.4868 247 XB70 0.6657 0.9676 0.7888 247 YF23 0.5490 0.7967 0.6501 246 Accuracy 0.6082 11353 Macro Avg 0.6804 0.6082 0.5787 11353 Weighted Avg 0.6803 0.6082 0.5787 11353
Potential Bias
- Bias in Training Data: Possible biases related to the data collection process.
- Limitations: Potential biases due to the nature of the dataset and model architecture.
Ethical Considerations
- Fairness: Address any concerns regarding fairness and potential bias in model predictions.
- Privacy: Describe any privacy considerations related to the model's deployment and use.
Model Usage Guidelines
- Recommended Use Cases: Military surveillance, security applications.
- Limitations: Clearly outline model limitations and potential failure scenarios.
- Legal and Ethical Considerations: Compliance with legal and ethical standards.
- Downloads last month
- 42
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.