WD 1.4 SwinV2 Tagger V2
| Property | Value |
|---|---|
| Parameter Count | 96.2M |
| License | Apache-2.0 |
| Framework Support | TF-Keras, ONNX, Safetensors |
| Tensor Type | F32 |
What is wd-v1-4-swinv2-tagger-v2?
The WD 1.4 SwinV2 Tagger V2 is an advanced image tagging model developed by SmilingWolf, specifically designed for comprehensive image analysis and tagging. This model represents a significant advancement in automated image tagging, trained on the extensive Danbooru dataset and capable of handling ratings, characters, and general tags with impressive accuracy.
Implementation Details
The model was trained using the SW-CV-ModelZoo framework, utilizing TPUs provided by the TRC program. It processes Danbooru images with specific ID criteria (modulo 0000-0899 for training, 0950-0999 for validation) and achieves an F1 score of 0.6854 at a threshold of 0.3771.
- Minimum requirement of 10 general tags per image
- Tags filtered to include only those with 600+ images
- Compatible with ONNXRuntime >= 1.17.0
- Supports batch inference
Core Capabilities
- Multi-framework support (TF-Keras, ONNX, Safetensors)
- Flexible batch processing
- Comprehensive tag coverage
- High accuracy in tag prediction
- Efficient processing with F32 tensor type
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its comprehensive tag coverage and efficient architecture, making it particularly suitable for large-scale image tagging tasks. The SwinV2 architecture provides excellent performance while maintaining reasonable computational requirements.
Q: What are the recommended use cases?
The model is ideal for automated image tagging systems, content organization, and image database management. It's particularly well-suited for applications requiring detailed image classification and tagging with multiple categories.
