EfficientNetV2-RW-S Model
| Property | Value |
|---|---|
| Parameter Count | 24.1M |
| GMACs | 4.9 |
| Image Size | Train: 288x288, Test: 384x384 |
| License | Apache-2.0 |
| Framework | PyTorch (timm) |
What is efficientnetv2_rw_s.ra2_in1k?
This is a specialized variant of EfficientNetV2 developed within the timm framework, designed for optimal performance in image classification tasks. The model implements the RandAugment (RA2) recipe, which was inspired by and evolved from original EfficientNet training methodologies, and was published as part of the 'ResNet Strikes Back' research.
Implementation Details
The model employs a sophisticated training approach using RMSProp optimizer with TF 1.0 behavior and EMA weight averaging. It features a step-based learning rate schedule with exponential decay and warmup periods.
- Architecture optimized for 21.4M activations
- Implements RandAugment RA2 recipe for enhanced training
- Supports flexible feature extraction and embedding generation
- Trained on ImageNet-1k dataset
Core Capabilities
- Image Classification with high accuracy
- Feature Map Extraction across multiple scales
- Image Embedding Generation
- Supports both training and inference modes
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its efficient architecture that balances speed and accuracy, incorporating the RA2 training recipe and optimized parameter count of 24.1M, making it particularly suitable for production deployments.
Q: What are the recommended use cases?
The model excels in image classification tasks, feature extraction for downstream tasks, and generating image embeddings. It's particularly well-suited for applications requiring both accuracy and computational efficiency.
