| Property | Value |
|---|---|
| Parameter Count | 315M |
| License | Apache-2.0 |
| Tensor Type | F32 |
| Framework | timm, ONNX, Safetensors |
What is wd-eva02-large-tagger-v3?
The WD EVA02-Large Tagger v3 is an advanced image tagging model specifically designed for comprehensive content classification. Developed by SmilingWolf using TPUs provided by the TRC program, this model represents a significant evolution in automated image tagging capabilities.
Implementation Details
The model was trained on Danbooru images with specific ID modulo ranges (0000-0899 for training, 0950-0999 for validation). It achieves a validation F1 score of 0.4772 at a threshold of 0.5296, demonstrating robust performance in tag prediction.
- Comprehensive dataset coverage up to February 2024
- Minimum threshold of 10 general tags per image
- Tags filtered to include only those with 600+ image examples
- Compatible with timm framework for easy integration
- Flexible ONNX implementation supporting batch inference
Core Capabilities
- Multi-category tagging support (ratings, characters, general tags)
- Batch processing capabilities
- High-performance inference with ONNX runtime (≥ 1.17.0)
- Seamless integration with multiple frameworks (JAX, timm, ONNX)
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its comprehensive training on a large-scale dataset with strict quality controls and multi-framework compatibility. The use of Macro-F1 for performance measurement ensures balanced evaluation across all tag categories.
Q: What are the recommended use cases?
The model is ideal for automated image tagging systems, content classification, and large-scale image database organization. It's particularly well-suited for applications requiring detailed anime and illustration content analysis.