MobileNetV3-Large 100 with RandAugment
| Property | Value |
|---|---|
| Parameter Count | 5.51M |
| Model Type | Image Classification |
| License | Apache-2.0 |
| Image Size | 224 x 224 |
| GMACs | 0.2 |
| Framework | PyTorch (timm) |
What is mobilenetv3_large_100.ra_in1k?
This is an optimized implementation of MobileNetV3-Large, trained on ImageNet-1k using the advanced RandAugment (RA) recipe. The model represents a significant advancement in mobile-first architecture design, combining efficiency with high performance. With just 5.51M parameters, it achieves impressive classification capabilities while maintaining a small footprint.
Implementation Details
The model utilizes the RandAugment training recipe, which was featured in the "ResNet Strikes Back" paper. It employs RMSProp optimization with TensorFlow 1.0 behavior and implements EMA weight averaging. The training process includes a step-based learning rate schedule with warmup periods.
- Efficient architecture with only 5.51M parameters
- Optimized for 224x224 input images
- Features TensorFlow-compatible RMSProp optimizer
- Implements advanced RandAugment data augmentation
Core Capabilities
- Image classification with state-of-the-art mobile performance
- Feature extraction for downstream tasks
- Embedding generation for transfer learning
- Efficient inference on mobile and edge devices
Frequently Asked Questions
Q: What makes this model unique?
This model stands out due to its implementation of the RandAugment training recipe, which significantly improves its performance compared to standard MobileNetV3 variants. It achieves an optimal balance between model size (5.51M parameters) and classification accuracy.
Q: What are the recommended use cases?
The model is ideal for mobile and edge deployment scenarios where resource constraints are important. It excels in real-time image classification tasks, feature extraction for custom vision applications, and as a backbone for transfer learning projects.
