Comparison between the HIWL and nine other galaxy classification works based on the Galaxy Zoo data set in literature. In this table, Overall val acc represents the highest overall accuracy on validation set, and Overall test acc represents the highest overall accuracy on test set. Num classes represents the number of classes to be divided. The accuracies of Reza (2021) and Lin et al. (2021) are based on the test set, and the others are based on the validation set.
| Method . | Overall . | Overall . | Num classes . |
|---|---|---|---|
| . | val acc . | test acc . | . |
| ANN (Reza 2021) | 98.2 per cent | 4 | |
| ResNet26 (Zhu et al. 2019) | 95.21 per cent | 5 | |
| SC-Net (Zhang et al. 2022) | 94.70 per cent | 5 | |
| NODE-ACA (Gupta et al. 2022) | 95.00 per cent | 5 | |
| Silva & Ventura (2019) | 94.01 per cent | 6 | |
| layered CNN (Goyal et al. 2020) | 88.33 per cent | 3 | |
| Jiménez et al. (2020) | 96.43 per cent | 2 | |
| ViT (Lin et al. (2021) | 81.21 per cent | 8 | |
| EfficientNet-B5 (Kalvankar et al. 2020) | 93.70 per cent | 7 | |
| HIWL | 97.22 per cent | 96.32 per cent | 5 |
| Method . | Overall . | Overall . | Num classes . |
|---|---|---|---|
| . | val acc . | test acc . | . |
| ANN (Reza 2021) | 98.2 per cent | 4 | |
| ResNet26 (Zhu et al. 2019) | 95.21 per cent | 5 | |
| SC-Net (Zhang et al. 2022) | 94.70 per cent | 5 | |
| NODE-ACA (Gupta et al. 2022) | 95.00 per cent | 5 | |
| Silva & Ventura (2019) | 94.01 per cent | 6 | |
| layered CNN (Goyal et al. 2020) | 88.33 per cent | 3 | |
| Jiménez et al. (2020) | 96.43 per cent | 2 | |
| ViT (Lin et al. (2021) | 81.21 per cent | 8 | |
| EfficientNet-B5 (Kalvankar et al. 2020) | 93.70 per cent | 7 | |
| HIWL | 97.22 per cent | 96.32 per cent | 5 |
Comparison between the HIWL and nine other galaxy classification works based on the Galaxy Zoo data set in literature. In this table, Overall val acc represents the highest overall accuracy on validation set, and Overall test acc represents the highest overall accuracy on test set. Num classes represents the number of classes to be divided. The accuracies of Reza (2021) and Lin et al. (2021) are based on the test set, and the others are based on the validation set.
| Method . | Overall . | Overall . | Num classes . |
|---|---|---|---|
| . | val acc . | test acc . | . |
| ANN (Reza 2021) | 98.2 per cent | 4 | |
| ResNet26 (Zhu et al. 2019) | 95.21 per cent | 5 | |
| SC-Net (Zhang et al. 2022) | 94.70 per cent | 5 | |
| NODE-ACA (Gupta et al. 2022) | 95.00 per cent | 5 | |
| Silva & Ventura (2019) | 94.01 per cent | 6 | |
| layered CNN (Goyal et al. 2020) | 88.33 per cent | 3 | |
| Jiménez et al. (2020) | 96.43 per cent | 2 | |
| ViT (Lin et al. (2021) | 81.21 per cent | 8 | |
| EfficientNet-B5 (Kalvankar et al. 2020) | 93.70 per cent | 7 | |
| HIWL | 97.22 per cent | 96.32 per cent | 5 |
| Method . | Overall . | Overall . | Num classes . |
|---|---|---|---|
| . | val acc . | test acc . | . |
| ANN (Reza 2021) | 98.2 per cent | 4 | |
| ResNet26 (Zhu et al. 2019) | 95.21 per cent | 5 | |
| SC-Net (Zhang et al. 2022) | 94.70 per cent | 5 | |
| NODE-ACA (Gupta et al. 2022) | 95.00 per cent | 5 | |
| Silva & Ventura (2019) | 94.01 per cent | 6 | |
| layered CNN (Goyal et al. 2020) | 88.33 per cent | 3 | |
| Jiménez et al. (2020) | 96.43 per cent | 2 | |
| ViT (Lin et al. (2021) | 81.21 per cent | 8 | |
| EfficientNet-B5 (Kalvankar et al. 2020) | 93.70 per cent | 7 | |
| HIWL | 97.22 per cent | 96.32 per cent | 5 |
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.
