Fig. 3 | Scientific Reports

Fig. 3

From: A spatiotemporal transformer with cross-frame encoding and trajectory-aware decoding for multi-target fish tracking

Fig. 3

The schematic of the Trajectory-Guided Decoder with Temporal Association Attention. This module leverages historical trajectory priors to generate queries, aggregates trajectory-related features through an attention mechanism, and fuses them with local appearance information before feeding them into the Transformer decoder, thereby enabling unified modeling of detection and association.

Back to article page