LCNet 050 RA2
| Property | Value |
|---|---|
| Parameter Count | 1.89M |
| Model Type | Image Classification |
| License | Apache-2.0 |
| Input Size | 224x224 |
| Paper | PP-LCNet Paper |
What is lcnet_050.ra2_in1k?
LCNet 050 RA2 is a lightweight convolutional neural network specifically designed for CPU deployment. It represents a balanced approach to efficient image classification, incorporating RandAugment (RA2) training techniques and achieving impressive performance with just 1.89M parameters.
Implementation Details
The model employs the RandAugment RA2 recipe, which evolved from EfficientNet's augmentation strategies. It uses RMSProp optimization with TF 1.0 behavior and implements exponential decay learning rate scheduling with warmup. The architecture is optimized for 224x224 input images and produces 1.3M activations.
- Trained on ImageNet-1k dataset
- Uses EMA weight averaging
- Implements step-based learning rate schedule
- Optimized for CPU inference
Core Capabilities
- Image classification with 1000 classes
- Feature extraction capabilities
- Image embedding generation
- Multi-scale feature map output
Frequently Asked Questions
Q: What makes this model unique?
LCNet 050 combines extremely lightweight architecture (1.89M parameters) with modern training techniques like RandAugment, making it particularly suitable for CPU deployment while maintaining competitive accuracy.
Q: What are the recommended use cases?
The model is ideal for resource-constrained environments, edge devices, and applications requiring real-time CPU inference. It's particularly suitable for mobile applications and embedded systems requiring image classification capabilities.
