FBNetV3-B Model
| Property | Value |
|---|---|
| Parameter Count | 8.64M |
| Model Type | Image Classification |
| License | Apache-2.0 |
| Dataset | ImageNet-1k |
| Framework | PyTorch (timm) |
What is fbnetv3_b.ra2_in1k?
FBNetV3-B is an efficient neural architecture designed through joint architecture-recipe search, specifically optimized for mobile and edge devices. This variant has been trained on ImageNet-1k using the advanced RandAugment (RA2) recipe, which was highlighted in the "ResNet Strikes Back" paper.
Implementation Details
The model employs a sophisticated training approach utilizing RMSProp optimizer with TensorFlow 1.0 behavior and EMA weight averaging. It features a step-based learning rate schedule with warmup and operates on 224x224 images during training, scaling to 256x256 for testing.
- Architecture: Efficient mobile-first design with 8.6M parameters
- Computational efficiency: Only 0.4 GMACs
- Memory footprint: 7.0M activations
- Advanced augmentation: Implements RA2 (RandAugment) recipe
Core Capabilities
- Image classification with 1000 classes (ImageNet)
- Feature extraction for downstream tasks
- Embedding generation for transfer learning
- Efficient inference on resource-constrained devices
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its efficient architecture derived through joint architecture-recipe search, making it particularly suitable for mobile deployments while maintaining competitive accuracy. The implementation of the RA2 training recipe further enhances its performance.
Q: What are the recommended use cases?
The model is ideal for mobile and edge device deployments requiring image classification capabilities. It's particularly well-suited for applications where computational resources are limited but accurate image classification is necessary.
