MobileNetV2 100 RA ImageNet-1k
| Property | Value |
|---|---|
| Parameters | 3.5M |
| GMACs | 0.3 |
| Image Size | 224 x 224 |
| Framework | timm |
| Dataset | ImageNet-1k |
| Paper | MobileNetV2: Inverted Residuals and Linear Bottlenecks |
What is mobilenetv2_100.ra_in1k?
MobileNetV2 100 RA is an efficient convolutional neural network designed for mobile and edge devices. This particular variant is trained using the RandAugment (RA) recipe, which enhances its performance through advanced data augmentation techniques. The model achieves an optimal balance between computational efficiency and accuracy, featuring just 3.5M parameters while maintaining robust classification capabilities.
Implementation Details
The model implements the MobileNetV2 architecture with RMSProp optimization (TF 1.0 behavior) and EMA weight averaging. It utilizes a step-based learning rate schedule with warmup and employs the RandAugment data augmentation strategy, as described in the "ResNet Strikes Back" paper.
- Activation Memory: 6.7M
- Efficient inverted residual structure
- Linear bottlenecks for feature preservation
- Optimized for mobile deployment
Core Capabilities
- Image classification with 1000 ImageNet classes
- Feature extraction with multiple resolution outputs
- Embedding generation for downstream tasks
- Efficient inference on resource-constrained devices
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its implementation of the RandAugment training recipe, which significantly improves its performance compared to standard MobileNetV2 variants. It achieves an excellent balance between model size (3.5M parameters) and computational efficiency (0.3 GMACs), making it ideal for mobile applications.
Q: What are the recommended use cases?
The model is particularly well-suited for mobile and edge computing applications requiring image classification, feature extraction, or embeddings generation. Its efficient architecture makes it ideal for real-time applications where computational resources are limited but accuracy cannot be significantly compromised.
