| Property | Value |
|---|---|
| Parameters | 11.7M |
| GMACs | 1.8 |
| Image Size | 224x224 |
| License | CC-BY-NC-4.0 |
| Paper | Billion-scale semi-supervised learning |
What is resnet18.fb_swsl_ig1b_ft_in1k?
This is a ResNet-18 architecture that has been pre-trained on Facebook's Instagram-1B dataset using semi-weakly supervised learning and subsequently fine-tuned on ImageNet-1k. The model represents an efficient implementation of the ResNet architecture, optimized for image classification tasks.
Implementation Details
The model implements a ResNet-B architecture with several key optimizations:
- ReLU activations for improved training stability
- Single layer 7x7 convolution with pooling for efficient feature extraction
- 1x1 convolution shortcut downsample for residual connections
- Trained using billion-scale semi-supervised learning approach
Core Capabilities
- Image Classification with 1000 classes
- Feature Map Extraction with multiple resolution outputs
- Image Embedding Generation
- Efficient inference with 1.8 GMACs computation requirement
Frequently Asked Questions
Q: What makes this model unique?
This model's uniqueness comes from its training approach - it was pre-trained on a massive dataset of 1 billion Instagram images using semi-weakly supervised learning, then fine-tuned on ImageNet. This gives it robust feature representations learned from real-world, diverse image data.
Q: What are the recommended use cases?
The model is well-suited for general image classification tasks, feature extraction for downstream tasks, and generating image embeddings. It offers a good balance between model size (11.7M parameters) and performance, making it suitable for production deployments where efficiency is important.