ConvNeXt Large MLP CLIP Model
| Property | Value |
|---|---|
| Parameter Count | 200M |
| Image Size | 320x320 |
| License | Apache 2.0 |
| Framework | PyTorch (timm) |
| Training Data | LAION-2B, ImageNet-12K, ImageNet-1K |
What is convnext_large_mlp.clip_laion2b_soup_ft_in12k_in1k_320?
This is a sophisticated ConvNeXt large model that combines CLIP pretraining on LAION-2B dataset with fine-tuning on ImageNet. The model represents a significant advancement in computer vision, achieving 87.97% top-1 accuracy on ImageNet validation while maintaining efficient inference with 70.2 GMACs.
Implementation Details
The model architecture is based on ConvNeXt with MLP adaptations, optimized for 320x320 resolution images. It leverages a multi-stage training approach: initial CLIP pretraining on LAION-2B, followed by fine-tuning on ImageNet-12K and final tuning on ImageNet-1K.
- Feature extraction capability with multi-scale outputs
- Efficient architecture with 200M parameters
- Optimized for both classification and embedding generation
- Supports various input resolutions with 320x320 being optimal
Core Capabilities
- Image Classification with high accuracy
- Feature Map Extraction at multiple scales
- Image Embedding Generation
- Transfer Learning Applications
Frequently Asked Questions
Q: What makes this model unique?
This model uniquely combines CLIP pretraining with hierarchical fine-tuning, resulting in robust visual representations that perform well across various tasks. The 320x320 resolution offers an optimal balance between accuracy and computational efficiency.
Q: What are the recommended use cases?
The model excels in image classification tasks, feature extraction for downstream applications, and generating image embeddings for similarity search or retrieval tasks. It's particularly well-suited for applications requiring high accuracy with reasonable computational requirements.
