ConvNeXt Nano Model
| Property | Value |
|---|---|
| Parameter Count | 15.6M |
| License | Apache 2.0 |
| Framework | PyTorch (timm) |
| Paper | A ConvNet for the 2020s |
What is convnext_nano.in12k_ft_in1k?
ConvNeXt Nano is a lightweight convolutional neural network designed for efficient image classification. Initially pre-trained on ImageNet-12k (a subset of ImageNet-22k with 11,821 classes) and fine-tuned on ImageNet-1k, it represents a modern approach to CNN architecture that combines efficiency with strong performance.
Implementation Details
The model features 15.6M parameters and requires 2.5 GMACs (billion multiply-accumulate operations) for inference. It operates on images with 224x224 resolution during training and 288x288 during testing, offering a good balance between computational efficiency and accuracy.
- Training performed on TPUs through Google's TRC program
- Fine-tuning completed on 8x GPU Lambda Labs cloud instances
- Achieves 82.282% top-1 accuracy on ImageNet-1k
- Processes 3926.52 samples per second with batch size 256
Core Capabilities
- Image classification with 1000 classes
- Feature map extraction at multiple scales
- Image embedding generation
- Efficient inference with F32 tensor type support
Frequently Asked Questions
Q: What makes this model unique?
This model represents an excellent balance between model size and performance, making it particularly suitable for resource-constrained applications while maintaining competitive accuracy. Its nano architecture is specifically optimized for efficiency while preserving the core benefits of the ConvNeXt architecture.
Q: What are the recommended use cases?
The model is well-suited for production environments where computational resources are limited but accurate image classification is required. It's particularly effective for mobile applications, edge devices, or scenarios requiring real-time inference with reasonable accuracy.
