ese_vovnet19b_dw.ra_in1k
| Property | Value |
|---|---|
| Parameter Count | 6.55M |
| Model Type | Image Classification / Feature Backbone |
| License | Apache-2.0 |
| Training Data | ImageNet-1k |
| Image Size | 224x224 (train) / 288x288 (test) |
What is ese_vovnet19b_dw.ra_in1k?
This is an energy-efficient implementation of the VoVNet architecture, specifically designed for optimal GPU computation. It's a variant that uses depthwise separable convolutions (dw) and was trained using the RandAugment (ra) recipe on ImageNet-1k. The model achieves a balance between computational efficiency and accuracy, requiring only 1.3 GMACs while maintaining strong performance.
Implementation Details
The model implements the VoVNet-v2 architecture with several optimizations. It uses energy-efficient spatial excitation (ESE) modules and features a lightweight design with 6.5M parameters. The architecture maintains 8.2M activations and employs different image sizes for training (224x224) and testing (288x288).
- Optimized backbone network for real-time object detection
- Implements depthwise separable convolutions for efficiency
- Trained using RandAugment augmentation strategy
- Supports both classification and feature extraction modes
Core Capabilities
- Image classification with 1000 classes (ImageNet)
- Feature map extraction at multiple scales
- Image embedding generation
- Real-time inference capability
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its energy-efficient design while maintaining competitive performance. The combination of ESE modules, depthwise separable convolutions, and RandAugment training makes it particularly suitable for resource-constrained environments.
Q: What are the recommended use cases?
The model is well-suited for real-time applications requiring efficient image classification or feature extraction. It's particularly appropriate for mobile and edge devices where computational resources are limited but real-time performance is necessary.
