Inception-v4 TensorFlow ImageNet Model
| Property | Value |
|---|---|
| Parameter Count | 42.7M |
| Model Type | Image Classification |
| License | Apache-2.0 |
| Paper | View Paper |
| Input Size | 299 x 299 |
| GMACs | 12.3 |
What is inception_v4.tf_in1k?
inception_v4.tf_in1k is a powerful image classification model that represents the fourth iteration of Google's Inception architecture. This particular implementation is a PyTorch port from the original TensorFlow model, maintained by the timm library. It combines deep learning techniques with efficient architecture design to achieve high accuracy on the ImageNet-1k dataset.
Implementation Details
The model features a sophisticated architecture with 42.7M parameters and requires 12.3 GMACs (billion multiply-accumulate operations) for inference. It processes images at 299x299 resolution and produces 1536-dimensional feature embeddings in its final layers.
- Optimized for ImageNet-1k classification tasks
- Supports feature map extraction with multiple output scales
- Includes pre-trained weights from TensorFlow conversion
- Implements both classification and feature extraction capabilities
Core Capabilities
- Image Classification with 1000-class ImageNet categories
- Feature Map Extraction at multiple scales
- Image Embedding Generation (1536-dimensional vectors)
- Flexible model usage with timm library integration
Frequently Asked Questions
Q: What makes this model unique?
This model represents a successful port of the TensorFlow Inception-v4 architecture to PyTorch, maintaining the original model's performance while providing the flexibility of the PyTorch ecosystem. Its architecture balances computational efficiency with high accuracy on image classification tasks.
Q: What are the recommended use cases?
The model is well-suited for large-scale image classification tasks, feature extraction for transfer learning, and as a backbone for more complex computer vision tasks. It's particularly effective when working with high-resolution images and when feature extraction at multiple scales is needed.
