TAT-VPR is a ternary-quantized transformer that brings dynamic accuracy-efficiency trade-offs to visual SLAM loop-closure. By fusing ternary weights with a learned activation-sparsity gate, the model can control computation by up to 40% at run-time without degrading performance (Recall@1). The proposed two-stage distillation pipeline preserves descriptor quality, letting it run on micro-UAV and embedded SLAM stacks while matching state-of-the-art localization accuracy.
View on arXiv@article{grainge2025_2505.16447,
title={ TAT-VPR: Ternary Adaptive Transformer for Dynamic and Efficient Visual Place Recognition },
author={ Oliver Grainge and Michael Milford and Indu Bodala and Sarvapali D. Ramchurn and Shoaib Ehsan },
journal={arXiv preprint arXiv:2505.16447},
year={ 2025 }
}