| Property | Value |
|---|---|
| Parameter Count | 66.7M |
| Image Size | 600x600 |
| License | Apache 2.0 |
| Framework | PyTorch (TimM) |
| Dataset | ImageNet-1K + JFT-300M |
What is tf_efficientnet_b7.ns_jft_in1k?
This model is an advanced implementation of EfficientNet B7 architecture, trained using the Noisy Student semi-supervised learning approach. Originally developed in TensorFlow and later ported to PyTorch by Ross Wightman, it combines training on ImageNet-1K with additional unlabeled data from JFT-300M to achieve superior performance.
Implementation Details
The model features 66.3M parameters and requires 38.3 GMACs for inference. It operates on 600x600 pixel images and produces 289.9M activations during processing. The architecture implements the compound scaling method from the original EfficientNet paper, optimized through Noisy Student training.
- Semi-supervised learning with Noisy Student training
- Optimized for high-resolution image processing
- Balanced efficiency and accuracy trade-off
- Compatible with both classification and feature extraction tasks
Core Capabilities
- Image Classification with 1000 classes
- Feature Map Extraction at multiple scales
- Image Embedding Generation
- Transfer Learning Applications
Frequently Asked Questions
Q: What makes this model unique?
This model stands out due to its Noisy Student training approach, which leverages both labeled ImageNet data and unlabeled JFT-300M data to improve performance. The B7 variant represents the largest scale in the EfficientNet family, optimized for scenarios requiring high accuracy.
Q: What are the recommended use cases?
The model excels in high-stakes image classification tasks, feature extraction for downstream tasks, and as a backbone for transfer learning. It's particularly suitable for applications requiring high accuracy and working with larger image resolutions.