Table 4 Explore various forms of exploration of polysemic contextual information on the VisDrone 2019 dataset.

From: End to end polysemantic cooperative mixed task trainer for UAV target detection

 

param#(M)

Flops

Accuracy (ALL-top-1)

Accuracy (ALL-top-5)

Static Polysemantic

17.0M

2.6G

31.37

48.13

Dynamic Polysemantic

20.2M

3.2G

33.13

48.73

Linear Polymerisation

20.2M

3.2G

33.15

48.74

PoT

22.1M

3.2G

33.83

48.77

  1. These include static-only polysemy, dynamic-only polysemy and linear aggregation only, as well as the full version of the PoT module (of which the main backbone architecture is PoT-50).
Back to article page