Develop

.gitignore

@@ -27,4 +27,31 @@ __pycache__/
 
 # OS
 .DS_Store
-Thumbs.db
+Thumbs.db
+
+# Outputs
+outputs/
+
+# Models
+*.pth
+*.pt
+*.ckpt
+
+# Logs
+*.log
+
+# Cache
+__pycache__/
+*.pyc
+
+# VSCode
+.vscode/
+
+# Conda
+.env
+.venv
+
+# Dataset
+data/raw/
+data/intermediate/
+data/processed/

PROJECT_STATUS.md

@@ -0,0 +1,68 @@
+# DeepVision AI v2.0
+
+## Sprint 1 – Foundation ✅
+
+- [x] Repository Recovery
+- [x] GitHub Repository
+- [x] Development Environment
+- [x] Configuration System
+
+---
+
+## Sprint 2 – Data Pipeline ✅
+
+- [x] Dataset Index Builder
+- [x] Dataset CSV Generator
+- [x] Sequence Dataset
+
+---
+
+## Sprint 3 – Training Infrastructure
+
+- [ ] Balanced Sampler
+- [ ] DataLoader
+- [ ] Trainer
+- [ ] Checkpoint Manager
+- [ ] Logger
+
+---
+
+## Sprint 4 – Model
+
+- [ ] ViT Integration
+- [ ] EfficientNet-BiLSTM
+- [ ] Model Factory
+
+---
+
+## Sprint 5 – Evaluation
+
+- [ ] Metrics
+- [ ] ROC Curve
+- [ ] Confusion Matrix
+- [ ] Cross Dataset Evaluation
+
+---
+
+## Sprint 6 – Explainability
+
+- [ ] Grad-CAM
+- [ ] Heatmap Overlay
+- [ ] Localization Metrics
+
+---
+
+## Sprint 7 – Deployment
+
+- [ ] Streamlit App
+- [ ] REST API
+- [ ] Docker
+
+---
+
+## Sprint 8 – Documentation
+
+- [ ] README
+- [ ] Architecture Diagram
+- [ ] Research Report
+- [ ] Portfolio Assets

README.md

@@ -0,0 +1,20 @@
+# DeepVision AI
+
+An Explainable Deepfake Detection Framework using Vision Transformers, Temporal Modeling, and Grad-CAM.
+
+## Status
+
+🚧 Under Active Development
+
+## Features
+
+- Vision Transformer based Deepfake Detection
+- Explainable AI using Grad-CAM
+- FF++ and Celeb-DF Support
+- Cross Dataset Evaluation
+- Streamlit Deployment
+- Research-oriented Modular Framework
+
+## Repository Structure
+
+See PROJECT_STATUS.md for the current development roadmap.

configs/default.yaml

@@ -0,0 +1,42 @@
+project:
+  name: DeepVision AI
+  version: 2.0
+
+paths:
+  raw_data: data/raw
+  processed_data: data/processed
+  sequences: data/intermediate/sequences
+  checkpoints: outputs/checkpoints
+  logs: outputs/logs
+  metrics: outputs/metrics
+  predictions: outputs/predictions
+
+dataset:
+  name: FFPP_CelebDF
+  image_size: 224
+  sequence_length: 30
+  fps: 5
+  batch_size: 8
+  num_workers: 4
+
+training:
+  epochs: 30
+  learning_rate: 0.0001
+  weight_decay: 0.0001
+  optimizer: AdamW
+  scheduler: CosineAnnealingLR
+
+model:
+  backbone: vit_b_16
+  pretrained: true
+  freeze_backbone: true
+  temporal_pooling: mean
+  num_classes: 2
+
+evaluation:
+  metrics:
+    - accuracy
+    - precision
+    - recall
+    - f1
+    - auc