TAT-VPR is a ternary-quantized vision transformer for visual place recognition that can dynamically skip up to 40% of computation at runtime with under 1% Recall@1 loss on one benchmark.
Boq: A place is worth a bag of learnable queries,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
TAT-VPR: Ternary Adaptive Transformer for Dynamic and Efficient Visual Place Recognition
TAT-VPR is a ternary-quantized vision transformer for visual place recognition that can dynamically skip up to 40% of computation at runtime with under 1% Recall@1 loss on one benchmark.