利用 PyTorch 在 CIFAR10 数据集上实现多种神经网络方法。
lr = 0.001, batch_size = 128, epoch = 300, GTX 2080 Ti
| model | best_acc |
|---|---|
| LeNet | 74.27% |
| VGG11 | 88.88% |
| VGG13 | 90.98% |
| VGG16 | 90.80% |
| ResNet18 | 92.19% |
| ResNet34 | 92.30% |
| ResNet50 | 91.39% |
| ResNet101 | 90.99% |
| ResNet152 | 90.70% |
| 模型 | 层数 | 论文链接 | 发表时间 | Google学术引用数(2019.12) |
|---|---|---|---|---|
| LeNet | 5 | Gradient-Based Learning Applied to Document Recognition | 1998 | 23110 |
| VGG | 11\13\16 | Very Deep Convolutional Networks for Large-Scale Image Recognition | ICLR 2015 | 31319 |
| ResNet | 18\34\50\101\152 | Deep residual learning for image recognition | CVPR2016 | 35470 |