Table 2 Learning from networks with different architectures and trained with different annotations.

From: Learning from algorithm-generated pseudo-annotations for detecting ants in videos

 

\(\tau _{base}:\hbox{U-Net}_{\mathrm{FE-1}}\)

\(\tau _{base}:\hbox{U-Net}_{\mathrm{FE-2}}\)

\(F_1\) score

AE

NN-Fit35

DE (ours)

AE

NN-Fit35

DE (ours)

\(\tau _{base}:\hbox{DCN}_{\mathrm{FE-1}}\)

0.7453

0.7538

0.7679

0.7331

0.7412

0.7562

\(\tau _{base}:\hbox{DCN}_{\mathrm{FE-2}}\)

0.7280

0.7352

0.7494

0.6232

0.6410

0.6605

  1. AE: Average Ensemble, DE: Deep Ensemble (Probabilistic NN-Fit). DCN was used for constructing the final model in NN-Fit and DE. Significant values are in bold.
Back to article page