Initial commit

.gitattributes

@@ -0,0 +1,35 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,75 @@
+---
+library_name: pytorch
+tags:
+  - video-generation
+  - flow-matching
+  - echocardiography
+  - medical-imaging
+pipeline_tag: other
+---
+
+# EchoLVFM — Weights
+
+One-step latent video flow matching for echocardiogram synthesis.
+
+- Paper: [arXiv:2603.13967](https://arxiv.org/abs/2603.13967)
+- Code: https://github.com/EngEmmanuel/EchoLVFM
+
+This repo holds **weights only**. The training + inference code lives in the
+`EchoLVFM` code repository. You need both to run the model.
+
+## Contents
+
+Three independent checkpoints, each in its own subfolder:
+
+| Subfolder | Flow | Inference | Notes |
+|-----------|------|-----------|-------|
+| `echolvfm_h1/` | RMMFlow | one-step | Adaptive-weighting exponent `h=1` in the training loss |
+| `echolvfm_h2/` | RMMFlow | one-step | Adaptive-weighting exponent `h=2` in the training loss |
+| `linear/`      | LinearFlow | multi-step ODE | Baseline for comparison |
+
+`h` is a **loss hyperparameter** (the exponent of the adaptive-weighting
+term), not a step count. Both RMMFlow variants are one-step generators —
+that's the defining property of RMMFlow.
+
+Each subfolder contains:
+- `model.safetensors` — the flow-level state dict (~293 MB).
+- `config.yaml` — minimal config to rebuild the UNet3D + flow wrapper.
+
+Subfolders load **independently**: a single call only downloads the
+requested variant's files (~293 MB), not the whole repo.
+
+## Loading
+
+```python
+from utils.hub import load_model_from_hub
+
+flow = load_model_from_hub(
+    "EngEmmanuel/EchoLVFM-Weights",
+    subfolder="echolvfm_h2",
+    device="cuda",
+)
+```
+
+You also need the paired VAE (`HReynaud/EchoFlow`, subfolder `vae`); see
+`vae/util.py::load_vae_and_processor` in the code repo.
+
+## Training data
+
+The underlying models were trained on VAE-encoded latents of the public
+[CAMUS](https://www.creatis.insa-lyon.fr/Challenge/camus/) dataset. Please
+respect the CAMUS dataset's license and citation requirements when using
+these weights.
+
+## Citation
+
+If you use EchoLVFM, please cite the paper:
+
+```bibtex
+@article{echolvfm2026,
+  title   = {EchoLVFM: One-Step Video Generation via Latent Flow Matching for Echocardiogram Synthesis},
+  author  = {Oladokun, Emmanuel and others},
+  journal = {arXiv preprint arXiv:2603.13967},
+  year    = {2026}
+}
+```

echolvfm_h1/config.yaml

@@ -0,0 +1,45 @@
+model:
+  type: unet
+  kwargs:
+    down_block_types:
+    - CrossAttnDownBlockSpatioTemporal
+    - CrossAttnDownBlockSpatioTemporal
+    - CrossAttnDownBlockSpatioTemporal
+    - CrossAttnDownBlockSpatioTemporal
+    up_block_types:
+    - UpBlockSpatioTemporal
+    - UpBlockSpatioTemporal
+    - UpBlockSpatioTemporal
+    - CrossAttnUpBlockSpatioTemporal
+    block_out_channels:
+    - 128
+    - 128
+    - 256
+    - 256
+    num_attention_heads:
+    - 4
+    - 4
+    - 8
+    - 8
+    cross_attention_dim: 1
+    layers_per_block: 2
+    transformer_layers_per_block:
+    - 1
+    - 1
+    - 2
+    - 2
+  init:
+    sample_size: 28
+    in_channels: 9
+    out_channels: 4
+    num_frames: 32
+flow:
+  type: mean
+  kwargs:
+    add_recon_loss: true
+    recon_loss_weight: 1.0
+    prob_default_flow_obj: 0.75
+    adaptive_loss_weight_p: 1
+    use_jvp_flash_attn: true
+  sample_kwargs:
+    steps: 1

echolvfm_h1/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:712a4ef12b0572aadfc4223d034a918289fb6bc4f64b7d7be9a930d8372ef3b8
+size 307285088

echolvfm_h2/config.yaml

@@ -0,0 +1,45 @@
+model:
+  type: unet
+  kwargs:
+    down_block_types:
+    - CrossAttnDownBlockSpatioTemporal
+    - CrossAttnDownBlockSpatioTemporal
+    - CrossAttnDownBlockSpatioTemporal
+    - CrossAttnDownBlockSpatioTemporal
+    up_block_types:
+    - UpBlockSpatioTemporal
+    - UpBlockSpatioTemporal
+    - UpBlockSpatioTemporal
+    - CrossAttnUpBlockSpatioTemporal
+    block_out_channels:
+    - 128
+    - 128
+    - 256
+    - 256
+    num_attention_heads:
+    - 4
+    - 4
+    - 8
+    - 8
+    cross_attention_dim: 1
+    layers_per_block: 2
+    transformer_layers_per_block:
+    - 1
+    - 1
+    - 2
+    - 2
+  init:
+    sample_size: 28
+    in_channels: 9
+    out_channels: 4
+    num_frames: 32
+flow:
+  type: mean
+  kwargs:
+    add_recon_loss: true
+    recon_loss_weight: 1.0
+    prob_default_flow_obj: 0.75
+    adaptive_loss_weight_p: 0.5
+    use_jvp_flash_attn: true
+  sample_kwargs:
+    steps: 1

echolvfm_h2/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ae294d97075765376afcf4b283f00c84de556ed715a2df9ddf62a895de01d38
+size 307285088

linear/config.yaml

@@ -0,0 +1,42 @@
+model:
+  type: unet
+  kwargs:
+    down_block_types:
+    - CrossAttnDownBlockSpatioTemporal
+    - CrossAttnDownBlockSpatioTemporal
+    - CrossAttnDownBlockSpatioTemporal
+    - CrossAttnDownBlockSpatioTemporal
+    up_block_types:
+    - UpBlockSpatioTemporal
+    - UpBlockSpatioTemporal
+    - UpBlockSpatioTemporal
+    - CrossAttnUpBlockSpatioTemporal
+    block_out_channels:
+    - 128
+    - 128
+    - 256
+    - 256
+    num_attention_heads:
+    - 4
+    - 4
+    - 8
+    - 8
+    cross_attention_dim: 1
+    layers_per_block: 2
+    transformer_layers_per_block:
+    - 1
+    - 1
+    - 2
+    - 2
+  init:
+    sample_size: 28
+    in_channels: 9
+    out_channels: 4
+    num_frames: 32
+flow:
+  type: linear
+  kwargs:
+    clip_values: null
+    clip_flow_values: null
+  sample_kwargs:
+    steps: 24

linear/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b84157df655090eddc491dd5dbf3b432ff207d7ae2f6c220ae0b13cc5a048344
+size 307285088