ResNet50 A1 ImageNet-1k Model
| Property | Value |
|---|---|
| Parameters | 25.6M |
| License | Apache-2.0 |
| Training Data | ImageNet-1k |
| Paper | ResNet Strikes Back |
| Architecture | ResNet-B variant |
What is resnet50.a1_in1k?
The resnet50.a1_in1k is a ResNet-50 model variant trained using the "ResNet Strikes Back" A1 training recipe on the ImageNet-1k dataset. This model represents a modern implementation of the classic ResNet architecture, incorporating latest training techniques for improved performance.
Implementation Details
This model implements the ResNet-B architecture with several key optimizations:
- Uses ReLU activation functions
- Implements a single 7x7 convolution layer with pooling
- Features 1x1 convolution shortcut downsample
- Trained using LAMB optimizer with BCE loss
- Employs cosine learning rate schedule with warmup
Core Capabilities
- Image classification with 1000 ImageNet classes
- Feature extraction capabilities
- Support for different input resolutions (224x224 training, 288x288 testing)
- Efficient inference with F32 tensor support
Frequently Asked Questions
Q: What makes this model unique?
This model implements the A1 training recipe from "ResNet Strikes Back", which modernizes the classic ResNet architecture with current best practices in training techniques. It achieves strong performance while maintaining the efficient ResNet-50 architecture.
Q: What are the recommended use cases?
The model is well-suited for image classification tasks, feature extraction, and as a backbone for more complex computer vision tasks. It offers a good balance between accuracy (81.22% top-1) and computational efficiency.
