Table 7 Cross-domain model performance

From: Neonatal pose estimation in the unaltered clinical environment with fusion of RGB, depth and IR images

Dataset

AP

AP50

AP75

RMSE

COCO (RGB)

0.698

0.832

0.738

–

COCO (RGB) → SyRIP

0.727

0.931

0.810

–

COCO (RGB) → Ours (RGB)

0.223

0.426

0.197

–

COCO (Gray)

0.682

0.800

0.729

–

COCO (Gray) → BabyPose

0.201

0.377

0.202

–

COCO (Gray) → Ours (Depth)

0.118

0.230

0.106

–

Ours (Depth)

0.765

0.944

0.850

7.11

Ours (Depth) → BabyPose

0.824

0.979

0.951

5.68

Ours (Depth) → COCO (Grayscale)

0.081

0.214

0.060

50.84

BabyPose*

–

–

–

9.06

Ours (RGB)

0.779

0.929

0.837

–

Ours (RGB) → SyRIP

0.383

0.639

0.395

–

Ours (RGB) → COCO (RGB)

0.351

0.588

0.360

–

SyRIP*

0.911

0.985

0.985

–

  1. A → B indicates a model is trained on A and tested on B. Models labelled * have scores taken directly from the respective authors’ original work. The RMSE is only included for depth models to allow comparison with BabyPose authors’ results.
Back to article page