coatnet_1¶
The coatnet_1 function registers and returns an instance of the CoAtNet model with a predefined configuration. It follows the CoAtNet-1 variant architecture, which balances convolutional and transformer layers for efficient image classification.
Total Parameters: 53,330,240
Function Signature¶
@register_model
def coatnet_1(num_classes: int = 1000, **kwargs) -> CoAtNet
Parameters¶
num_classes (int, optional): The number of output classes for classification. Default is 1000.
kwargs (dict, optional): Additional arguments passed to the CoAtNet constructor.
Configuration¶
The coatnet_1 model follows the configuration from the original CoAtNet paper. The number of blocks and channels in each stage are summarized in the table below:
Stage |
Block Type |
Blocks |
Channels |
|---|---|---|---|
Stage 1 |
Convolution |
2 |
96 |
Stage 2 |
Convolution |
6 |
192 |
Stage 3 |
Transformer |
14 |
384 |
Stage 4 |
Transformer |
2 |
768 |
Examples¶
Basic Example
import lucid.models as models
# Instantiate coatnet_1 with default 1000 classes
model = models.coatnet_1(num_classes=1000)
# Input tensor with shape (1, 3, 224, 224)
input_ = lucid.random.randn(1, 3, 224, 224)
# Perform forward pass
output = model(input_)
print(output.shape) # Shape: (1, 1000)
Custom Number of Classes
# Instantiate coatnet_1 for 10-class classification
model = models.coatnet_1(num_classes=10)
input_ = lucid.random.randn(1, 3, 224, 224)
output = model(input_)
print(output.shape) # Shape: (1, 10)