| Property | Value |
|---|---|
| Parameter Count | 98M |
| License | Apache-2.0 |
| Tensor Type | F32 |
| Framework | timm, ONNX, Safetensors |
What is wd-swinv2-tagger-v3?
WD SwinV2 Tagger v3 is a sophisticated image tagging model specifically designed for anime and manga content analysis. Built using the SwinV2 architecture, this model represents a significant advancement in automated content tagging, trained on an extensive Danbooru dataset up to image ID 7220105.
Implementation Details
The model utilizes a frequency-based loss scaling approach to address class imbalance issues, trained on Danbooru images with IDs modulo 0000-0899 and validated on images with IDs modulo 0950-0999. It achieves an impressive F1 score of 0.4541 at a threshold of 0.2653, showing improvement over previous versions.
- Comprehensive training on images with 10+ general tags
- Tag filtering threshold of 600+ images
- Compatible with timm and ONNX runtime (>= 1.17.0)
- Flexible batch processing capability
Core Capabilities
- Ratings classification
- Character recognition
- General tag generation
- Batch inference support
- Cross-platform compatibility
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its optimized performance through frequency-based loss scaling and extensive dataset coverage, making it particularly effective for anime/manga content tagging. The flexible architecture supports multiple frameworks and batch processing capabilities.
Q: What are the recommended use cases?
The model is ideal for automated content tagging in anime/manga collections, content moderation systems, and large-scale media organization tasks. It's particularly effective when integrated into systems requiring accurate character recognition and general content classification.