tf_efficientnet_b0.ns_jft_in1k
| Property | Value |
|---|---|
| Parameter Count | 5.33M |
| Model Type | Image Classification |
| License | Apache 2.0 |
| Framework | PyTorch (timm) |
| Input Size | 224x224 |
| Training Data | ImageNet-1k + JFT-300M |
What is tf_efficientnet_b0.ns_jft_in1k?
This model is a PyTorch implementation of the EfficientNet-B0 architecture, enhanced through Noisy Student training on both ImageNet-1k and the massive JFT-300M dataset. Originally developed in TensorFlow by the paper authors and later ported to PyTorch by Ross Wightman, it represents an optimal balance between model size and accuracy for image classification tasks.
Implementation Details
The model employs compound scaling principles from the EfficientNet family, optimizing width, depth, and resolution simultaneously. With 5.3M parameters and 0.4 GMACs, it achieves impressive efficiency while maintaining strong performance. The architecture utilizes mobile inverted bottleneck convolutions and squeeze-and-excitation blocks for effective feature extraction.
- Trained using Noisy Student semi-supervised learning
- Optimized for 224x224 input images
- Features 6.7M activations
- Supports both classification and feature extraction modes
Core Capabilities
- Image classification with 1000 classes (ImageNet)
- Feature map extraction at multiple scales
- Image embedding generation
- Transfer learning foundation
Frequently Asked Questions
Q: What makes this model unique?
This model combines the efficient architecture of EfficientNet-B0 with Noisy Student training on an extensive dataset (JFT-300M), resulting in robust performance while maintaining a small parameter footprint.
Q: What are the recommended use cases?
The model excels in general image classification tasks, transfer learning applications, and as a feature extractor for downstream computer vision tasks. It's particularly suitable for mobile and edge deployments due to its efficient design.
