Delete README.md

README.md

@@ -1,86 +0,0 @@
----
-license: apache-2.0
----
-## Vision Transformer (ViT) Models for Digital Forensics
-
-This repository provides Vision Transformer (ViT) models fine-tuned to detect manipulated (fake) versus authentic (real) image frames extracted from the FaceForensics++ dataset. The models were trained using Intel® ARC GPU (XPU-enabled) and optimized for binary image classification in digital forensics workflows.
-
----
-
-## 🧠 Models Trained
-
-| Model Name                 | Pretrained On         | Patch Size | Parameters |
-|---------------------------|-----------------------|------------|------------|
-| `vit_tiny_patch16_224`    | ImageNet-21k          | 16×16      | ~5.7M      |
-| `vit_tiny_patch32_224`    | ImageNet-21k          | 32×32      | ~5.6M      |
-| `vit_small_patch32_224`   | AugReg + IN21k + IN1k | 32×32      | ~22M       |
-| `vit_large_patch16_224`   | ImageNet-21k          | 16×16      | ~304M      |
-| `vit_large_patch32_224`   | ImageNet-21k          | 32×32      | ~304M      |
-
----
-
-## 🗂️ Dataset
-
-- **Name**: DeepFake Detection (DFD)  
-- **Source**: [Kaggle DFD Dataset] 
-- **Classes**: `real`, `fake`  
-- **Input**: Extracted video frames resized to 224×224 RGB images  
-- **Preprocessing**:
-  - Resizing and normalization using `torchvision.transforms`
-  - Structured into `train/real`, `train/fake`, `val/real`, `val/fake`
-
----
-
-## ⚙️ Hardware & Environment
-
-- **Accelerator**: Intel® ARC GPU (XPU via Intel Extension for PyTorch)
-- **Frameworks**:
-  - PyTorch 2.7.0 + XPU backend
-  - torchvision 0.22.0
-  - timm for pretrained ViT models
-- **OS**: Windows 11
-- **Memory Consideration**: `vit_huge_patch14_224` requires large GPU memory; tested on Intel ARC A770 16GB and NPU Boost
-
----
-
-## ✅ Use Case: Deepfake Frame Detection
-
-These models are designed to identify manipulated media content at the frame level. Use cases include:
-
-- 🔍 Video forensics
-- 🎞️ Deepfake screening and flagging pipelines
-- 🧪 Data validation for machine learning datasets
-- 📡 Real-time frame-level media authentication
-
-They are well-suited for deployment in digital forensics, content moderation, and research scenarios where image authenticity is critical.
-
----
-
-## 📊 Results 
-
-| Model | Train Accuracy | Validation Accuracy |
-|-------|----------------|---------------------|
-| [vit_large_patch16_224](https://huggingface.co/pranav2711/VisionTransformerDigitalForensics/blob/main/vit_large_patch16_224.pth) | **94.89%** | **91.22%** |
-| [vit_large_patch32_224](https://huggingface.co/pranav2711/VisionTransformerDigitalForensics/blob/main/vit_large_patch32_224.pth) | 91.31% | 89.23% |
-| [vit_tiny_patch16_224](https://huggingface.co/pranav2711/VisionTransformerDigitalForensics/blob/main/vit_tiny_patch16_224.pth) | 92.41% | 89.20% |
-| [vit_small_patch32_224](https://huggingface.co/pranav2711/VisionTransformerDigitalForensics/blob/main/vit_small_patch32_224.pth) | 91.38% | 88.29% |
-| [vit_small_patch16_224](https://huggingface.co/pranav2711/VisionTransformerDigitalForensics/blob/main/vit_small_patch32_224.pth) | 80.67% | 81.25% |
-| [vit_base_patch16_224](http://huggingface.co/pranav2711/VisionTransformerDigitalForensics/blob/main/vit_base_patch16_224.pth) | 90.65% | 85.36% |
-| [vit_base_patch32_224](https://huggingface.co/pranav2711/VisionTransformerDigitalForensics/blob/main/vit_base_patch32_224.pth) | 79.54% | 79.54% |
-
-**vit_large_patch16_224**
-This model achieved the highest validation accuracy of 91.22% with strong training stability and generalization. It is recommended as the final model for deployment or downstream tasks.
-
----
-
-## 📄 License
-
-This model is licensed under the [CreativeML OpenRAIL-M License](https://huggingface.co/spaces/CompVis/stable-diffusion-license).  
-It allows for responsible research and commercial use, but **strictly prohibits**:
-
-- Harassment, surveillance, or profiling.
-- Generating misleading or harmful content (e.g., deepfakes for impersonation).
-- Use in political campaigns or autonomous weapons.
-
-Please read the license carefully before using the model.
-