Table 3 Summary of volume estimation accuracy across portion sizes for the modified texture foods dataset across our proposed EDFN and EDFN-D.
Dataset | Segmentation accuracy | Intake accuracy | Volume estimation accuracy (mL) | ||||
|---|---|---|---|---|---|---|---|
Portion | GSA | FSA | 2D % intake error | 3D % intake error | Mean absolute error | Mean error | Volume intake error |
EDFN (no-depth-refinement) | |||||||
P1 | 0.996 (0.003) | 0.970 (0.066) | 0.0 (0.0) | 0.0 (0.0) | 3.3 (3.4) | 2.0 (4.3) | 0.0 (0.0) |
P2 | 0.994 (0.006) | 0.965 (0.089) | − 30.2 (17.2) | − 0.8 (8.0) | 3.0 (3.1) | 2.1 (3.8) | − 0.1 (3.3) |
P3 | 0.992 (0.009) | 0.941 (0.125) | − 50.1 (23.5) | − 0.9 (5.3) | 2.9 (2.9) | 1.9 (3.6) | 0.1 (3.7) |
P4 | 0.987 (0.013) | 0.912 (0.176) | − 68.0 (35.3) | − 1.1 (5.6) | 3.0 (3.2) | 1.4 (4.2) | 0.6 (4.5) |
P5 | 0.982 (0.014) | 0.821 (0.254) | − 71.9 (43.4) | − 1.0 (3.8) | 2.0 (2.6) | 1.1 (3.1) | 0.9 (4.6) |
EDFN-D (depth-refined) | |||||||
P1 | 0.996 (0.004) | 0.945 (0.074) | 0.0 (0.0) | 0.0 (0.0) | 2.6 (3.3) | 0.1 (4.3) | 0.0 (0.0) |
P2 | 0.994 (0.010) | 0.908 (0.132) | − 22.8 (17.5) | 0.3 (8.7) | 1.9 (3.0) | 0.0 (3.6) | 0.1 (3.3) |
P3 | 0.992 (0.013) | 0.824 (0.162) | − 30.6 (17.5) | 1.4 (6.3) | 2.3 (3.5) | − 0.5 (4.1) | 0.7 (4.2) |
P4 | 0.987 (0.018) | 0.640 (0.275) | − 25.2 (20.1) | 2.8 (6.6) | 2.9 (3.8) | − 1.7 (4.5) | 1.8 (4.4) |
P5 | 0.986 (0.016) | 0.157 (0.262) | − 3.9 (16.2) | 2.3 (3.5) | 1.7 (2.2) | − 1.5 (2.4) | 1.6 (4.0) |
