Spaces:

---

mcosarinsky

/

CheXmask-U

like 1

Running

CheXmask-U is a deep learning model for landmark-based chest X-ray segmentation with node-wise uncertainty estimation. The Space hosts an interactive demo that allows users to upload chest X-rays and visualize anatomical landmarks along with predictive uncertainty.

The model uses a hybrid graph-convolutional architecture with variational latent sampling, which requires multiple stochastic forward passes per image to compute uncertainty. GPU ensures fast and responsive inference for interactive demos and batch processing of large datasets.

The Space enables researchers and the community to explore uncertainty-aware anatomical segmentation in chest X-rays and facilitates reproducibility and safe deployment of landmark-based segmentation methods.