Fig. 3 | Scientific Reports

Fig. 3

From: Dynamic adaptive synergistic attention network for visible-infrared person re-identification

Fig. 3

AKSA overview. The input feature maps are reshaped into a sequence and processed in parallel by a Channel Enhancement Branch and a Position Enhancement Branch. Their outputs are concatenated and passed through a lightweight gate network that adaptively fuses the two streams into a single feature. A subsequent global aggregation produces the weight tensor used by the following channel switching.

Back to article page