Models¶
Collection of (generative) models for ultrasound imaging.
zea contains a collection of models for various tasks, all located in the zea.models package.
See the following dropdown for a list of available models:
Available models
zea.models.echonet.EchoNetDynamic: A model for left ventricle segmentation.zea.models.carotid_segmenter.CarotidSegmenter: A model for carotid artery segmentation.zea.models.echonetlvh.EchoNetLVH: A model for left ventricle hypertrophy segmentation.zea.models.unet.UNet: A simple U-Net implementation.zea.models.lpips.LPIPS: A model implementing the perceptual similarity metric.zea.models.taesd.TinyAutoencoder: A tiny autoencoder model for image compression.zea.models.regional_quality.MobileNetv2RegionalQuality: A scoring model for myocardial regions in apical views.zea.models.lv_segmentation.AugmentedCamusSeg: A nnU-Net based left ventricle and myocardium segmentation model.zea.models.speckle2self.Speckle2Self: A self-supervised speckle reduction model for ultrasound images.
Presets for these models can be found in zea.models.presets. Presets are pre-trained weights for the models, which can be used to initialize the models for inference or further training. Each model class has a presets attribute that lists the available presets for that model. We store the presets on Hugging Face Hub, and they are downloaded automatically when loading a model with a preset.
To use these models, you can import them directly from the zea.models module and load the pretrained weights using the from_preset() method. For example:
>>> from zea.models.unet import UNet
>>> model = UNet.from_preset("unet-echonet-inpainter")
You can list all available presets using the presets attribute:
>>> from zea.models.unet import UNet
>>> presets = list(UNet.presets.keys())
>>> print(f"Available built-in zea presets for UNet: {presets}")
Available built-in zea presets for UNet: ['unet-echonet-inpainter']
Generative models¶
In addition to regular models, zea provides generative models for tasks such as image
generation, inpainting, and denoising. The key difference is that generative models have
sampling methods implemented. There are two base classes:
GenerativeModel for classical models (e.g. a Gaussian
mixture model) and DeepGenerativeModel for
neural-network-based models — the latter also inherits from
BaseModel, adding Keras features like weight saving and preset
loading. Both expose the following methods:
fit()for training the model on datasample()for generating new samples from the learned distributionposterior_sample()for drawing samples from the posterior given measurementslog_density()for computing the log-probability of data under the model
See the following dropdown for a list of available generative models:
Available models
zea.models.diffusion.DiffusionModel: A deep generative diffusion model for ultrasound image generation.zea.models.flow_matching.FlowMatchingModel: A flow matching generative model for ultrasound image generation.zea.models.gmm.GaussianMixtureModel: A Gaussian Mixture Model.zea.models.hvae.HierarchicalVAE: A hierarchical variational autoencoder for ultrasound image generation.
An example of how to use the zea.models.diffusion.DiffusionModel is shown below:
>>> from zea.models.diffusion import DiffusionModel
>>> model = DiffusionModel.from_preset("diffusion-echonet-dynamic")
>>> samples = model.sample(n_samples=4)
Adding a new model¶
New models are welcome! Please follow the Contributing guide for the general workflow (forking, branches, pull requests, etc.). The steps below walk you through what is specific to adding a model.
Create a new file in
zea/models/for your model, e.g.zea/models/mymodel.py.Add a model class that inherits from
zea.models.base.BaseModel. For generative models, useGenerativeModelorDeepGenerativeModelas the base class. Implement thecallmethod.Upload the pretrained weights to our Hugging Face. The expected files are a
config.jsonand amodel.weights.h5. See the Keras documentation for how to save these. You can drag and drop the files directly on the Hugging Face website.Tip
Alternate saving methods are also possible. See
zea.models.echonet.EchoNetfor an example — in that case you need to implement acustom_load_weightsmethod in your model class.Add a preset for your model in
zea.models.presets. Presets let you register multiple sets of weights for the same model architecture.In your model file, import the presets module and call
register_presetswith your model class to activate the presets.Import your model in
zea/models/__init__.pyto make it part of the package.
Adding non-Keras (custom) models¶
The recommended approach for any model is to implement it as a native Keras 3 model. This gives you backend-agnostic execution (JAX, TensorFlow, PyTorch) and the full preset/weight-loading infrastructure for free.
For models originally trained in PyTorch, the typical workflow is:
Vendor the architecture — copy the PyTorch network code into your module (e.g. inside a
_build_torch_classes()helper that importstorchlazily so that PyTorch is only required for weight conversion, not inference).Implement the Keras architecture — write
keras.layers.Layersubclasses that replicate each block. Key API differences to handle:Padding for stride-2 Conv2D: Keras
padding='same'is asymmetric forstride > 1; useZeroPadding2D(1) + Conv2D(padding='valid')to match PyTorch’s symmetricpadding=1.ConvTranspose: Keras
Conv2DTranspose(padding='valid')gives the full output; cropx[:, 1:, 1:, :](NHWC) to reproduce PyTorch’sConvTranspose2d(padding=1, output_padding=1)alignment.InstanceNorm: use
GroupNormalization(groups=C, scale=False, center=False, epsilon=1e-5)forInstanceNorm2d(affine=False).Weight axes: Conv2D —
(2,3,1,0); Conv2DTranspose —(2,3,1,0)(same permutation, different semantics).Input format: Keras defaults to channels-last (NHWC); transpose NCHW → NHWC in
call()and back before returning.
Write a weight-loading helper — a function that maps PyTorch state-dict keys to the Keras layer tree and calls
layer.set_weights([...]).Add
from_pth(path)classmethod — wraps the weight loader for convenient local testing.Optionally add an ONNX fallback — for environments that have
onnxruntimebut nottorch, you can keep afrom_onnx(path)classmethod and an_onnx_sessattribute; overridecall()to dispatch to the ONNX path when the session is set.Follow steps 3-6 from the standard guide above for HF upload, presets, and registration.
See zea.models.speckle2self for a complete worked example of this
pattern (native Keras + PyTorch weight loading + optional ONNX fallback).