dpn107.mx_in1k
| Property | Value |
|---|---|
| Parameter Count | 87.1M |
| Model Type | Image Classification |
| Architecture | Dual Path Network |
| License | Apache-2.0 |
| Paper | Dual Path Networks |
What is dpn107.mx_in1k?
The dpn107.mx_in1k is a sophisticated image classification model based on the Dual Path Networks architecture. Originally trained in MXNet and ported to PyTorch by Ross Wightman, this model represents a clever fusion of ResNet and DenseNet architectures, offering 87.1M parameters and requiring 18.4 GMACs for inference.
Implementation Details
This implementation features a complex architecture that processes 224x224 images through multiple feature extraction stages. The model produces feature maps at various resolutions, with the final layer generating 2688 channels at 7x7 resolution before classification.
- Trained on ImageNet-1k dataset
- Supports various inference modes including classification and feature extraction
- Provides 33.5M activation points
- Implements dual path architecture for optimal feature reuse
Core Capabilities
- Image classification with 1000-class ImageNet categories
- Feature map extraction at multiple scales
- Embedding generation for transfer learning
- Flexible preprocessing with model-specific transforms
Frequently Asked Questions
Q: What makes this model unique?
The DPN107 uniquely combines the advantages of ResNet's feature reuse and DenseNet's new feature exploration, creating a highly efficient architecture that balances model capacity with computational requirements.
Q: What are the recommended use cases?
This model is ideal for high-accuracy image classification tasks, transfer learning applications, and as a backbone for more complex computer vision tasks like object detection or segmentation.
