RegNetY-032 Model
| Property | Value |
|---|---|
| Parameters | 19.4M |
| GMACs | 3.2 |
| Image Size (Train) | 224 x 224 |
| Image Size (Test) | 288 x 288 |
| Paper | Designing Network Design Spaces |
What is regnety_032.ra_in1k?
RegNetY-032 is a state-of-the-art image classification model trained on ImageNet-1k, implementing the RegNet architecture with several timm-specific enhancements. It represents an efficient balance between computational cost and accuracy, achieving 82.74% top-1 accuracy at 288px resolution.
Implementation Details
This implementation by timm includes several advanced features that distinguish it from standard RegNet implementations:
- Stochastic depth for improved regularization
- Gradient checkpointing to optimize memory usage
- Layer-wise learning rate decay
- Configurable output stride with dilation support
- Flexible activation and normalization layers
- Pre-activation bottleneck block option
Core Capabilities
- Image classification with 1000 classes
- Feature extraction for downstream tasks
- Flexible input resolution support
- Memory-efficient training and inference
Frequently Asked Questions
Q: What makes this model unique?
This model combines the efficient RegNetY architecture with timm-specific optimizations, offering a strong balance between computational efficiency (3.2 GMACs) and accuracy. It's particularly notable for its enhanced implementation features not found in standard RegNet models.
Q: What are the recommended use cases?
The model is well-suited for image classification tasks, feature extraction, and as a backbone for computer vision tasks requiring efficient processing with good accuracy. It performs particularly well in scenarios where a balance between computational resources and accuracy is needed.
