MobileNetV4 Conv Small
| Property | Value |
|---|---|
| Parameter Count | 3.8M |
| License | Apache 2.0 |
| Framework | PyTorch (timm) |
| Paper | MobileNetV4 Paper |
| Top-1 Accuracy | 74.6% |
What is mobilenetv4_conv_small.e2400_r224_in1k?
This is a compact variant of the MobileNetV4 architecture, specifically designed for mobile and edge computing applications. Trained by Ross Wightman using the timm framework, it represents a balance between model size and performance, achieving 74.6% top-1 accuracy on ImageNet-1k while maintaining a lightweight footprint of only 3.8M parameters.
Implementation Details
The model features a modern convolutional architecture optimized for mobile deployment, with training conducted over 2400 epochs on 224x224 resolution images. It utilizes efficient convolution operations and achieves 0.2 GMACs with 2.0M activations.
- Trained on ImageNet-1k dataset
- Supports both 224x224 (training) and 256x256 (inference) image sizes
- Implements the latest MobileNetV4 architecture innovations
- Provides feature extraction capabilities with multiple resolution outputs
Core Capabilities
- Image Classification: Performs 1000-class ImageNet classification
- Feature Extraction: Supports multi-scale feature map extraction
- Embedding Generation: Can generate image embeddings for downstream tasks
- Mobile-Optimized: Designed for efficient mobile deployment
Frequently Asked Questions
Q: What makes this model unique?
This model represents one of the first available implementations of MobileNetV4, with no official TensorFlow weights currently released. It offers an excellent balance between model size and performance, making it ideal for mobile applications.
Q: What are the recommended use cases?
The model is particularly well-suited for mobile and edge device deployment where computational resources are limited. It's effective for image classification tasks, feature extraction, and as a backbone for more complex computer vision applications.
