Image Classification on CIFAR-10 (test) (Accuracy, Size, Latency, Speedup)
81.42AccuracyHybrid (Prune50% → QAT → KD)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Hybrid (Prune50% → QAT → KD)Backbone=VGG-16-BN2026.04 | 81.42 | 7.03 | 0.99 | 2.65 | |
| Hybrid (Prune50% → KD → QAT)Backbone=VGG-16-BN2026.04 | 80.9 | 7.03 | 0.98 | 2.67 | |
| Prune (50%)Backbone=VGG-16-BN2026.04 | 80.55 | 28.11 | 2.81 | 0.93 | |
| Prune (50%)Backbone=ResNet-182026.04 | 80.38 | 26.97 | 2.55 | 0.96 | |
| Hybrid (QAT → Prune50% → KD)Backbone=VGG-16-BN2026.04 | 80.3 | 7.03 | 0.99 | 2.65 | |
| Prune (30%)Backbone=VGG-16-BN2026.04 | 80.11 | 39.33 | 2.64 | 0.99 | |
| Prune (30%)Backbone=ResNet-182026.04 | 79.84 | 33.13 | 2.43 | 1.01 | |
| Hybrid (Prune50% → QAT → KD)Backbone=ResNet-182026.04 | 79.62 | 6.74 | 1 | 2.45 | |
| BaselineBackbone=VGG-16-BN2026.04 | 79.38 | 56.18 | 2.62 | 1 | |
| Hybrid (Prune50% → KD → QAT)Backbone=ResNet-182026.04 | 79 | 6.74 | 1.02 | 2.4 | |
| Hybrid (QAT → KD → Prune50%)Backbone=VGG-16-BN2026.04 | 78.8 | 7.03 | 1 | 2.62 | |
| KDBackbone=VGG-16-BN2026.04 | 78.57 | 56.18 | 2.69 | 0.97 | |
| Hybrid (QAT → Prune50% → KD)Backbone=ResNet-182026.04 | 78.5 | 6.74 | 1 | 2.45 | |
| BaselineBackbone=ResNet-182026.04 | 78.37 | 42.65 | 2.45 | 1 | |
| QAT (INT8)Backbone=ResNet-182026.04 | 77.42 | 10.66 | 0.99 | 2.47 | |
| QAT (INT8)Backbone=VGG-16-BN2026.04 | 77.23 | 14.05 | 1.01 | 2.59 | |
| Hybrid (QAT → KD → Prune50%)Backbone=ResNet-182026.04 | 76.6 | 6.74 | 1.01 | 2.43 | |
| KDBackbone=ResNet-182026.04 | 76.33 | 42.65 | 2.49 | 0.98 |