---
license: apache-2.0
---

### Hierarchical Variational Autoencoder (HVAE)

Generative model for 2D B-mode Ultrasound Images. 
This model can be used for unconditional sampling as well as inpainting tasks.

Current architecture in this folder is only trained on resolution `256x256x3` (3 video frames per input/output).

🌽 The `zea` implementation is a snippet, adapted from the [source](https://github.com/swpenninga/hvae).

The weights that are currently present can be read as follows:
- `lvh`: trained on `EchonetLVH` dataset
- `ur`: encoder retrained with UniformRandomLines subsampling [`zea.agent`](https://github.com/tue-bmd/zea/blob/main/zea/agent/selection.py)
- `ge`: encoder retrained with GreedyEntropy subsampling [`zea.agent`](https://github.com/tue-bmd/zea/blob/main/zea/agent/selection.py)
- `24`: 24 lines/columns used when subsampling

📚 A usage tutorial can be found on [Collab](https://colab.research.google.com/github/tue-bmd/zea/blob/main/docs/source/notebooks/models/hvae_model_example.ipynb)