This is also just a practice for learning pytorch. It is the second practice. The first one is here So here is the network:
class CNN(nn.Module):
def __init__(self):
super().__init__()
self.cnn = nn.Sequential(
# Input 3x48x48
nn.Conv2d(in_channels=3, out_channels=64, kernel_size=3, padding=1),
nn.BatchNorm2d(64),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2),
# Input 64x24x24
nn.Conv2d(in_channels=64, out_channels=128, kernel_size=3, padding=1),
nn.BatchNorm2d(128),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2),
# Input 128x12x12
nn.Conv2d(in_channels=128, out_channels=256, kernel_size=3, padding=1),
nn.BatchNorm2d(256),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2),
# Input 256x6x6
nn.AdaptiveAvgPool2d((1, 1)),
nn.Flatten(),
nn.Linear(256, 128),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Linear(128, 8),
)
def forward(self, x):
return self.cnn(x)
This model's accuracy after 50 epochs will reach around 60%
Epoch 50/50 ----------------------------------------
Batch 0/252 | Loss: 1.1548 | Samples/sec: 2865 | LR: 1.0e-04
Batch 50/252 | Loss: 1.1075 | Samples/sec: 146 | LR: 1.0e-04
Batch 100/252 | Loss: 1.1158 | Samples/sec: 146 | LR: 1.0e-04
Batch 150/252 | Loss: 1.1156 | Samples/sec: 147 | LR: 1.0e-04
Batch 200/252 | Loss: 1.1179 | Samples/sec: 146 | LR: 1.0e-04
Batch 250/252 | Loss: 1.1191 | Samples/sec: 149 | LR: 1.0e-04
Test Results | Accuracy: 59.7% | Avg loss: 1.0893
Done!
There are tensorboard logs in
logs
├── events.out.tfevents.1751519047.bear.1056600.0
└── events.out.tfevents.1751519544.bear.1064911.0
https://www.kaggle.com/datasets/mstjebashazida/affectnet
Grade 9 2025/07/03