Comparison of the test results before and after combining the method HIWL with the classical model. The table contains 11 comparison models such as Dieleman, ResNet, Vision Transoformer, etc. Avg acc represents the average accuracy of 10 runs for each model, Avg acc (with HIWL) represents the average accuracy of 10 runs for each model after incorporating HIWL, and Promotion represents the difference between the accuracy of the model after incorporating HIWL and the original model.
| Model . | Avg acc . | Avg acc . | Promotion . |
|---|---|---|---|
| . | . | (with HIWL) . | . |
| Dieleman (Dieleman et al. 2015) | 0.9337 | 0.9400 | 0.0063 |
| ResNet26 (Zhu et al. 2019) | 0.9074 | 0.9147 | 0.0073 |
| VGG16 (Simonyan & Zisserman 2015) | 0.9431 | 0.9469 | 0.0038 |
| GoogleNet (Szegedy et al. 2015) | 0.9480 | 0.9507 | 0.0027 |
| ResNet34 (He et al. 2016) | 0.9507 | 0.9597 | 0.0090 |
| ResNet50 (He et al. 2016) | 0.9469 | 0.9497 | 0.0028 |
| EfficientNet-B0 (Tan & Le 2019) | 0.9521 | 0.9577 | 0.0056 |
| EfficientNet-B1 (Tan & Le 2019) | 0.9542 | 0.9612 | 0.0070 |
| EfficientNet-B2 (Tan & Le 2019) | 0.9503 | 0.9580 | 0.0077 |
| Vision Transformer (Dosovitskiy et al. 2021) | 0.9264 | 0.9451 | 0.0187 |
| Model . | Avg acc . | Avg acc . | Promotion . |
|---|---|---|---|
| . | . | (with HIWL) . | . |
| Dieleman (Dieleman et al. 2015) | 0.9337 | 0.9400 | 0.0063 |
| ResNet26 (Zhu et al. 2019) | 0.9074 | 0.9147 | 0.0073 |
| VGG16 (Simonyan & Zisserman 2015) | 0.9431 | 0.9469 | 0.0038 |
| GoogleNet (Szegedy et al. 2015) | 0.9480 | 0.9507 | 0.0027 |
| ResNet34 (He et al. 2016) | 0.9507 | 0.9597 | 0.0090 |
| ResNet50 (He et al. 2016) | 0.9469 | 0.9497 | 0.0028 |
| EfficientNet-B0 (Tan & Le 2019) | 0.9521 | 0.9577 | 0.0056 |
| EfficientNet-B1 (Tan & Le 2019) | 0.9542 | 0.9612 | 0.0070 |
| EfficientNet-B2 (Tan & Le 2019) | 0.9503 | 0.9580 | 0.0077 |
| Vision Transformer (Dosovitskiy et al. 2021) | 0.9264 | 0.9451 | 0.0187 |
Comparison of the test results before and after combining the method HIWL with the classical model. The table contains 11 comparison models such as Dieleman, ResNet, Vision Transoformer, etc. Avg acc represents the average accuracy of 10 runs for each model, Avg acc (with HIWL) represents the average accuracy of 10 runs for each model after incorporating HIWL, and Promotion represents the difference between the accuracy of the model after incorporating HIWL and the original model.
| Model . | Avg acc . | Avg acc . | Promotion . |
|---|---|---|---|
| . | . | (with HIWL) . | . |
| Dieleman (Dieleman et al. 2015) | 0.9337 | 0.9400 | 0.0063 |
| ResNet26 (Zhu et al. 2019) | 0.9074 | 0.9147 | 0.0073 |
| VGG16 (Simonyan & Zisserman 2015) | 0.9431 | 0.9469 | 0.0038 |
| GoogleNet (Szegedy et al. 2015) | 0.9480 | 0.9507 | 0.0027 |
| ResNet34 (He et al. 2016) | 0.9507 | 0.9597 | 0.0090 |
| ResNet50 (He et al. 2016) | 0.9469 | 0.9497 | 0.0028 |
| EfficientNet-B0 (Tan & Le 2019) | 0.9521 | 0.9577 | 0.0056 |
| EfficientNet-B1 (Tan & Le 2019) | 0.9542 | 0.9612 | 0.0070 |
| EfficientNet-B2 (Tan & Le 2019) | 0.9503 | 0.9580 | 0.0077 |
| Vision Transformer (Dosovitskiy et al. 2021) | 0.9264 | 0.9451 | 0.0187 |
| Model . | Avg acc . | Avg acc . | Promotion . |
|---|---|---|---|
| . | . | (with HIWL) . | . |
| Dieleman (Dieleman et al. 2015) | 0.9337 | 0.9400 | 0.0063 |
| ResNet26 (Zhu et al. 2019) | 0.9074 | 0.9147 | 0.0073 |
| VGG16 (Simonyan & Zisserman 2015) | 0.9431 | 0.9469 | 0.0038 |
| GoogleNet (Szegedy et al. 2015) | 0.9480 | 0.9507 | 0.0027 |
| ResNet34 (He et al. 2016) | 0.9507 | 0.9597 | 0.0090 |
| ResNet50 (He et al. 2016) | 0.9469 | 0.9497 | 0.0028 |
| EfficientNet-B0 (Tan & Le 2019) | 0.9521 | 0.9577 | 0.0056 |
| EfficientNet-B1 (Tan & Le 2019) | 0.9542 | 0.9612 | 0.0070 |
| EfficientNet-B2 (Tan & Le 2019) | 0.9503 | 0.9580 | 0.0077 |
| Vision Transformer (Dosovitskiy et al. 2021) | 0.9264 | 0.9451 | 0.0187 |
This PDF is available to Subscribers Only
View Article Abstract & Purchase OptionsFor full access to this pdf, sign in to an existing account, or purchase an annual subscription.
