Table 5 Comparison of VQEDTL Variants with baseline DLMs.
Metric | ISONet | Classical ResNet50 | VQ ResNet50_10 | Classical MobileNetV2 | VQ MobileNetV2_10 | Classical Inception ResNetV2 | VQ Inception ResNetV2_10 | Classical DenseNet201 | VQ DenseNet201_10 |
|---|---|---|---|---|---|---|---|---|---|
Total params | 3,278,371 | 23,859,333 | 23,606,433 | 2,431,301 | 2,273,633 | 54,542,821 | 54,353,409 | 18,577,221 | 18,340,193 |
Training params | 3,278,371 | 23,805,829 | 18,337 | 2,396,805 | 15,265 | 54,481,893 | 16,289 | 18,347,781 | 17,825 |
FLOPs | 65, 56,742 | 47,718,66 | 36,674 | 48,62,602 | 30,530 | 10,89,63,786 | 32,578 | 3,66,95,562 | 35,650 |
Inference time (s) | 2.09 | 1.19 | 0.295 | 0.95 | 0.259 | 3.499 | 0.655 | 3.58 | 0.589 |
Validation Acc(%) | 94.7 | 95.6 | 99.08 | 85.69 | 95.32 | 84.77 | 88.9 | 92.48 | 97.52 |
