| Property | Value |
|---|---|
| Parameter Count | 11.7M |
| License | Apache-2.0 |
| Training Data | ImageNet-1k |
| Paper | ResNet strikes back: An improved training procedure in timm |
What is resnet18.a1_in1k?
ResNet18.a1_in1k is a ResNet-B architecture model trained on ImageNet-1k using the A1 recipe from the "ResNet Strikes Back" paper. It features a lightweight architecture with 11.7M parameters while maintaining good performance for practical applications.
Implementation Details
The model implements the ResNet architecture with several key optimizations:
- ReLU activations for improved training stability
- Single layer 7x7 convolution with pooling
- 1x1 convolution shortcut downsample
- LAMB optimizer with BCE loss during training
- Cosine LR schedule with warmup
Core Capabilities
- Image Classification with 1000 classes
- Feature Map Extraction across multiple scales
- Image Embedding Generation
- Efficient inference with 1.8 GMACs
Frequently Asked Questions
Q: What makes this model unique?
This model represents an optimal balance between model size and performance, using the A1 training recipe that includes modern training techniques like LAMB optimizer and cosine learning rate scheduling. It's particularly suitable for resource-constrained applications.
Q: What are the recommended use cases?
The model is well-suited for general image classification tasks, feature extraction, and as a backbone for transfer learning. It performs best with images of size 224x224 pixels during training and can handle 288x288 during inference.