Awesome
C++ implementation of papers Mobile Vision Transformer-based Visual Object Tracking (BMVC2023) and Separable Self and Mixed Attention Transformers for Efficient Object Tracking (WACV2024)
The original Pytorch implementations is available in these links: MVT and SMAT.
Why C++ implementation?
- Improve the model inference speed (fps)
- Single vs Multi-thread throughput analysis
- Lower memory footprint compared to Python
- Subsequent deployment on Andorid devices using JNI