TABLE 6.
Comparison of the performance of different image-based food-recognition systems on publicly available food datasets1
| Dataset | Reference | CNN-based approach | ACC, % | Mean Average Precision, % |
|---|---|---|---|---|
| PFID | (118) | – | 50.45 | — |
| (76) | + | 70.13 | ||
| UEC-Food 100 | (66) | – | 60.5 | |
| (48) | + | 75 | ||
| (88) | + | 81.45 | ||
| (37) | + | 82.12 | ||
| (77) | + | 60.9 | ||
| (122) | + | 76.3 | ||
| (102) | – | 82.38 | ||
| (118) | – | 60.50 | ||
| (126) | + | 86.51 | ||
| (92) | + | 49.19 | ||
| (129) | + | 89.58 | ||
| (46) | + | 81.0 | ||
| (93) | + | 84.52 | ||
| (117) | + | 17.9 | ||
| (132) | + | — | 17.5 | |
| (134) | + | 60.90 | ||
| UEC-Food 256 | (17) | + | 63.16 | |
| (88) | + | 76.17 | ||
| (125) | + | — | 31.7 | |
| (126) | + | 78.60 | ||
| (129) | + | 83.15 | ||
| (46) | + | 72.0 | ||
| (93) | + | 77.20 | ||
| (116) | + | 93.0 | ||
| (132) | + | — | 10.5 | |
| Food-101 | (88) | + | 88.28 | |
| (120) | + | 72.11 | ||
| (121) | + | 58.65 | ||
| (122) | + | 77.4 | ||
| (123) | + | 71.12 | ||
| (126) | + | 87.96 | ||
| (127) | + | 86.97 | ||
| (119) | – | 68.29 | ||
| (128) | + | 81.65 | ||
| (129) | + | 90.27 | ||
| (46) | + | 80.0 | ||
| (93) | + | 84.28 | ||
| (131) | + | 55.3 | ||
| (133) | + | 79.86 | ||
| (135) | + | 80.0 | ||
| (86) | + | 64.98 | ||
| (136) | + | 74.02 | ||
| UNIMIB 2016 | (6) | + | 78.0 | |
| (99) | – | 95.9 | ||
| (100) | – | 94.5 | ||
| (98) | – | 93.9 | ||
| (110) | + | 86.39 | ||
| (96) | – | 96.27 | ||
| (130) | + | 77.5 | ||
| VIREO Food-172 | (37) | + | 82.06 | |
| (131) | + | 75.1 | ||
| (117) | + | 24.2 | ||
| Madima 2017 | (103) | + | 93.33 | |
| (97) | + | 57.1 | ||
| FOOD-5K | (133) | + | 99.0 | |
| (86) | + | 98.8 | ||
| (38) | + | 99.2 | ||
| Food-11 | (133) | + | 89.33 | |
| (86) | + | 91.34 | ||
| (38) | + | 83.6 |
1
ACC, accuracy; CNN, convolutional neural network; PFID, Pittsburgh Fast-food Image Dataset; –, not using a CNN-based approach; +, using a CNN-based approach.