Update README.md

README.md

@@ -1,20 +1,11 @@
 ---
+---
 license: mit
 language:
 - en
 base_model:
 - google/vit-base-patch16-224
 pipeline_tag: image-classification
-tags:
-- deepfake
-- fakeimages
-- detector
-- fake
-- vision-transformer
-- vit
-- image-classification
-- computer-vision
-- deep-learning
 model_name: ViT Deepfake Detector
 model_creator: Hamza Sohail, Ayaan Mohammed, Shadab Karim, Kirti Dhir
 model_type: vision-transformer
@@ -22,43 +13,31 @@ datasets:
 - faceforensics++
 - celeb-df
 - dfdc
-- custom-generated
 library_name: transformers
 library_version: "4.40.0"
 inference: true
 model_description: |
-  This model is a fine-tuned version of Google's `vit-base-patch16-224` Vision Transformer, trained specifically for the binary classification task of detecting deepfake images. It outputs probabilities indicating whether a given image is real or fake.
-
-  The model was trained using a combination of real and manipulated images sourced from the FaceForensics++, Celeb-DF, and DFDC datasets, along with additional synthetic samples. It leverages the ViT architecture's ability to capture spatial and contextual features across image patches for effective fake content detection.
-
-  The primary application of this model is in fake image detection, digital media integrity validation, and social platform moderation tools.
+  A fine-tuned Vision Transformer (`vit-base-patch16-224`) for classifying real vs. fake images. Trained on FaceForensics++, Celeb-DF, DFDC, and custom samples. Outputs real/fake probabilities for input images.
 
 training_details: |
-  - Base Model: google/vit-base-patch16-224
   - Epochs: 10
   - Optimizer: AdamW
-  - Loss: CrossEntropyLoss
-  - Learning rate: 5e-5
-  - Scheduler: CosineAnnealingLR
+  - Loss: CrossEntropy
+  - LR: 5e-5
   - Batch size: 32
-  - Framework: PyTorch with Hugging Face Transformers
-  - Hardware: Trained using Tesla T4 GPU
+  - GPU: Tesla T4
 
 evaluation: |
-  The model was evaluated on a stratified test set of 10,000 images from multiple sources, achieving:
-  - Accuracy: 95.7
-  - Precision: 95.7%
-  - Recall: 95.7%
-  - F1-score: 95.7%
-
-  Confusion matrix and ROC curves were generated to analyze misclassifications and detection confidence.
+  Evaluated on 10,000 images:
+  - Accuracy: 95.7%
+  - Precision/Recall/F1: 95.7%
 
 intended_uses: |
-  This model is intended for:
-  - Automated detection of manipulated or deepfake images in social media content.
-  - Research in digital forensics and AI ethics.
-  - Educational purposes for understanding the application of Vision Transformers.
-
-  **Limitations:** This model may not generalize to unseen manipulation techniques not present in the training datasets. It is not intended for use in real-time legal or security-critical applications without additional verification mechanisms.
-
-example_usage: 
+  For fake image detection in research, moderation, and education. Not for legal/critical decisions without further verification.
+tags:
+- deepfake
+- fakeimages
+- detector
+- vit
+- computer-vision
+- deep-learning