TAT-VPR: Ternary Adaptive Transformer for Dynamic and Efficient Visual Place Recognition
This work addresses efficiency challenges for micro-UAV and embedded SLAM systems, offering a dynamic trade-off solution, though it is incremental as it builds on existing transformer and quantization methods.
The paper tackles the problem of balancing accuracy and efficiency in visual place recognition for SLAM loop-closure by introducing TAT-VPR, a ternary-quantized transformer with dynamic computation control, achieving up to 40% reduction in computation without performance degradation in Recall@1.
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.