📱 Offline Image-to-Text Translation App (iOS)

A Swift-based mobile app for iOS that performs offline image-to-text translation entirely using locally deployed models.

This project compares CoreML pipelines with custom multimodal architectures to extract and translate text from images without any server-side components or internet access.

---

🚀 Key Features

• 📸 Camera & Photo Support – Select or capture an image from your device.
• 🧠 On-Device Text Detection – Extracts text using local models with no external API calls.
• 🌍 Offline Translation – Translates text to a language of your choice using small, efficient translation models.
• 🧬 Multimodal Architecture – Improves robustness and quality in noisy or low-light images.

---

🔍 Core Findings

• ✅ Better Accuracy: Using a locally deployed multimodal model has improves both text recognition and translation quality compared to an Apple-only, traditional CoreML pipeline.

---

🧰 Tech Stack

• Language: Swift
• Frameworks: UIKit, CoreML, Vision, Natural Language
• ML Models: Custom multimodal transformer-based models deployed with CoreML
• Platform: iOS 16+

---

📦 Future Enhancements

• Add tokenization and other support for low-resource languages
• Add in-place translation with bounding boxes
• Batch processing for multiple images
• Evaluate quantized models for faster inference, particularly on older devices

---