cx-tissue-seg
| Property | Value |
|---|---|
| Author | conflux-xyz |
| Architecture | UNet with MobileNet-v3 encoder |
| Performance | 0.93 mIoU on test set |
| Model URL | Hugging Face |
What is cx-tissue-seg?
cx-tissue-seg is a specialized deep learning model designed for binary segmentation of tissue in H&E pathology slides. The model stands out for its efficiency, capable of processing entire slides in under 1 second on standard CPU hardware. It employs a UNet architecture with a MobileNet-v3 encoder, specifically optimized for resource-constrained environments.
Implementation Details
The model is trained on 512x512 pixel patches from thumbnail images at 40 microns per pixel (MPP). It's available in multiple formats including PyTorch (.pth), SafeTensors, and ONNX, with an additional quantized int8 version for enhanced efficiency. The implementation uses the timm/mobilenetv3_small_100 as the encoder backbone, ensuring optimal performance while maintaining computational efficiency.
- Trained on manually curated H&E pathology slides
- Outputs logits for binary classification (tissue vs. background)
- Includes tiled inference framework for full image processing
- Multiple format support (PyTorch, ONNX, SafeTensors)
Core Capabilities
- Fast tissue segmentation (< 1 second per slide on CPU)
- High accuracy with 0.93 mIoU on test data
- Efficient processing of large whole-slide images
- Support for both CPU and GPU inference
- Integrated tiling and stitching functionality
Frequently Asked Questions
Q: What makes this model unique?
The model's primary strength lies in its combination of speed and accuracy, processing whole slides in under a second while maintaining 0.93 mIoU accuracy. Its efficient architecture makes it particularly suitable for resource-constrained environments.
Q: What are the recommended use cases?
The model is ideal for automated tissue segmentation in digital pathology workflows, particularly when processing H&E stained slides. It's especially useful in high-throughput scenarios where rapid processing is crucial, or in environments with limited computational resources.
