Table 4 Inference efficiency of interactive segmentation methods on the HECKTOR 2021 dataset, reporting peak VRAM usage and mean inference times (with standard deviation) on an NVIDIA RTX 3080 GPU and Intel i5-12600K CPU.

From: Interactive 3D segmentation for primary gross tumor volume in oropharyngeal cancer

Method

VRAM (GB)

NVIDIA RTX 3080 (10 GB)

Intel i5-12600K CPU

Inference time (s)

Inference time (s)

2S-ICR (ours)

2.06

\(0.08\pm 0.00\)

\(1.62\pm 0.05\)

DeepGrow12,13

1.86

\(0.08\pm 0.01\)

\(1.63\pm 0.05\)

DeepEdit-2513

1.86

\(0.08\pm 0.01\)

\(1.63\pm 0.05\)

DeepEdit-5013

1.86

\(0.08\pm 0.01\)

\(1.63\pm 0.05\)

U-Net15

1.88

\(0.08\pm 0.00\)

\(1.63\pm 0.05\)

  1. U-Net15 is the non-interactive baseline.