zeahub/flowmatching-echonetlvh

FlowMatching Cardiac Ultrasound model

A Generative diffusion model for 2D B-mode cardiac ultrasound images.

This model can be used for unconditional sampling as well as posterior sampling and inpainting tasks.

• The model is trained on EchonetLVH at a resolution of 256x256
• The three folders 1ch, 3ch, 12ch contain a 1-frame, 3-frame and 12-frame model respectively, all with the same unet architecture, but producing different length sequences.
• The three folders 1ch-dit, 3ch-dit, 7ch-dit are similarly named, with a Diffusion Transformer (DiT) model, instead of a UNet.

---

Simply import the models with:

from zea.models.flow_matching import FlowMatchingModel
model_1ch = FlowMatchingModel.from_preset("flowmatching-echonetlvh")
model_3ch = FlowMatchingModel.from_preset("flowmatching-echonetlvh-3ch")
model_12ch = FlowMatchingModel.from_preset("flowmatching-echonetlvh-12ch")

model_1ch = FlowMatchingModel.from_preset("flowmatching-echonetlvh-dit")
model_3ch = FlowMatchingModel.from_preset("flowmatching-echonetlvh-dit-3ch")
model_12ch = FlowMatchingModel.from_preset("flowmatching-echonetlvh-dit-12ch")