Work
On-device computer vision
AI.Reverie
- Role
- Machine Learning Engineer
- Team
- Computer Vision Team
- When
- 2020
Real-time object detection and instance segmentation on iOS: PyTorch models converted to LibTorch and run on-device through a Swift + AVFoundation capture pipeline.
What I did
- Lead engineer on a deep-learning + iOS app for real-time object detection and instance segmentation.
- Improved accuracy with dropout, batch normalisation, and training on custom synthetic data.
- Converted models to run on-device via LibTorch; wrote the Objective-C++ bridge, hand-rolled NMS, and the full AVFoundation capture pipeline. CoreMotion integration plotted phone-path trajectory.
- Productionised the library and ETL pipeline with custom PyTorch DataLoaders and dataset mappers.
Full write-up in progress. The bullets above are the short version.
Under the hood
Stack.
Deep learning
- Dropout, batch normalisation, hyperparameter tuning
- Training on custom synthetic data
- Custom PyTorch DataLoaders and dataset mappers
- TorchScript model conversion for mobile
On-device mobile ML
- iOS object detection + instance segmentation
- Swift + AVFoundation capture pipeline
- LibTorch via Objective-C++ bridge
- Device motion logging for trajectory analysis