ResNet34 A1 ImageNet-1k Model
| Property | Value |
|---|---|
| Parameters | 21.8M |
| License | Apache-2.0 |
| Architecture | ResNet-B |
| Training Data | ImageNet-1k |
| Top-1 Accuracy | 77.92% |
| Paper | ResNet Strikes Back |
What is resnet34.a1_in1k?
ResNet34.a1_in1k is a ResNet-B architecture model trained on the ImageNet-1k dataset using the modern "ResNet strikes back" A1 training recipe. It features 21.8M parameters and achieves 77.92% top-1 accuracy, making it an efficient choice for image classification tasks.
Implementation Details
The model implements the ResNet-B architecture with several key optimizations:
- ReLU activations throughout the network
- Single 7x7 convolution layer with pooling at the input
- 1x1 convolution shortcut downsample paths
- Trained using LAMB optimizer with BCE loss
- Cosine learning rate schedule with warmup
Core Capabilities
- Image classification on 1000 ImageNet classes
- Feature extraction with multiple resolution options
- Efficient inference with 3.7 GMACs compute requirement
- Flexible input sizes (224x224 training, 288x288 testing)
Frequently Asked Questions
Q: What makes this model unique?
This model implements the modern A1 training recipe from "ResNet Strikes Back", which demonstrates that ResNet architectures can achieve competitive performance when trained with current best practices. It offers a good balance between model size and accuracy.
Q: What are the recommended use cases?
The model is well-suited for general image classification tasks, transfer learning applications, and feature extraction. It performs particularly well when moderate model size and good accuracy are required, making it suitable for production deployments with resource constraints.
