upload weights

1SCM_v2/config.yaml

@@ -0,0 +1,59 @@
+unets:
+  unet1:
+    dim: 64
+    num_resnet_blocks: 2
+    dim_mults:
+    - 1
+    - 2
+    - 4
+    max_text_len: 1
+    layer_attns: false
+    layer_cross_attns:
+    - false
+    - false
+    - true
+    cond_images_channels: 3
+    attend_at_middle: false
+imagen:
+  condition_on_text: true
+  image_sizes:
+  - 112
+  text_embed_dim: 1
+  num_sample_steps: 64
+  temporal_downsample_factor:
+  - 1
+  lowres_sample_noise_level: 0.2
+  sigma_min: 0.002
+  sigma_max: 80
+  sigma_data: 0.25
+  rho: 7
+  P_mean: -1.2
+  P_std: 1.2
+  S_churn: 160
+  S_tmin: 0.05
+  S_tmax: 50
+  S_noise: 1.003
+trainer:
+  split_batches: false
+  lr: 0.0005
+  dl_tuple_output_keywords_names:
+  - images
+  - text_embeds
+  - cond_images
+dataset:
+  data_path: /path/to/EchoNet-Dynamic
+  deactivate_cache: false
+  fps: 8
+  duration: 2.0
+  grayscale: false
+dataloader:
+  batch_size: 8
+  num_workers: 8
+wandb:
+  project: EchoDiffusion
+  # entity: your-login
+checkpoint:
+  path: /path/to/checkpoints
+  batch_size: 4
+  cond_scale: 5.0
+  save_every_x_it: 5000

1SCM_v2/merged.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:724eb5f0580198979f174dd41ee3747ee8c8cc73ebcbdff973f7682ab00b1e74
+size 300968361

2SCM_v6/config.yaml

@@ -0,0 +1,83 @@
+unets:
+  unet1:
+    dim: 64
+    num_resnet_blocks: 2
+    dim_mults:
+    - 1
+    - 2
+    - 4
+    max_text_len: 1
+    layer_attns: false
+    layer_cross_attns:
+    - false
+    - false
+    - true
+    cond_images_channels: 3
+  unet2:
+    dim: 64
+    num_resnet_blocks: 2
+    dim_mults:
+    - 1
+    - 2
+    - 4
+    max_text_len: 1
+    layer_attns: false
+    layer_cross_attns:
+    - false
+    - false
+    - true
+    memory_efficient: true
+    cond_images_channels: 3
+imagen:
+  elucidated: true
+  condition_on_text: true
+  image_sizes:
+  - 56
+  - 112
+  text_embed_dim: 1
+  num_sample_steps:
+  - 32
+  - 64
+  random_crop_sizes:
+  - null
+  - 56
+  temporal_downsample_factor:
+  - 4
+  - 1
+  sigma_min: 0.002
+  sigma_max: 80
+  sigma_data: 0.25
+  rho: 7
+  P_mean: -1.2
+  P_std: 1.2
+  S_churn:
+  - 80
+  - 160
+  S_tmin: 0.05
+  S_tmax: 50
+  S_noise: 1.003
+  resize_mode: trilinear
+trainer:
+  split_batches: false
+  lr: 0.0005
+  dl_tuple_output_keywords_names:
+  - images
+  - text_embeds
+  - cond_images
+dataset:
+  data_path: /path/to/EchoNet-Dynamic
+  deactivate_cache: false
+  fps: 32
+  duration: 2.0
+  grayscale: false
+dataloader:
+  batch_size: 8
+  num_workers: 8
+wandb:
+  project: EchoDiffusion
+  # entity: your-login
+checkpoint:
+  path: /path/to/checkpoints
+  batch_size: 4
+  cond_scale: 5.0
+  save_every_x_it: 5000

2SCM_v6/merged.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:daf9c964854734145651c69540e8befdad3ed741b8cefa4ca85fd11cf6d2f353
+size 755800751

4SCM_v2/config.yaml

@@ -0,0 +1,119 @@
+unets:
+  unet1:
+    dim: 64
+    num_resnet_blocks: 2
+    dim_mults:
+    - 1
+    - 2
+    - 4
+    max_text_len: 1
+    layer_attns: false
+    layer_cross_attns:
+    - false
+    - false
+    - true
+    cond_images_channels: 3
+  unet2:
+    dim: 64
+    num_resnet_blocks: 2
+    dim_mults:
+    - 1
+    - 2
+    - 4
+    max_text_len: 1
+    layer_attns: false
+    layer_cross_attns:
+    - false
+    - false
+    - true
+    memory_efficient: true
+    cond_images_channels: 3
+  unet3:
+    dim: 64
+    num_resnet_blocks: 2
+    dim_mults:
+    - 1
+    - 2
+    - 4
+    max_text_len: 1
+    layer_attns: false
+    layer_cross_attns:
+    - false
+    - false
+    - true
+    memory_efficient: true
+    cond_images_channels: 3
+  unet4:
+    dim: 64
+    num_resnet_blocks: 2
+    dim_mults:
+    - 1
+    - 2
+    - 4
+    max_text_len: 1
+    layer_attns: false
+    layer_cross_attns: false
+    cond_images_channels: 3
+    attend_at_middle: false
+imagen:
+  condition_on_text: true
+  image_sizes:
+  - 56
+  - 56
+  - 56
+  - 112
+  text_embed_dim: 1
+  num_sample_steps:
+  - 32
+  - 32
+  - 32
+  - 64
+  random_crop_sizes:
+  - null
+  - null
+  - null
+  - 56
+  temporal_downsample_factor:
+  - 4
+  - 2
+  - 1
+  - 1
+  lowres_sample_noise_level: 0.2
+  sigma_min: 0.002
+  sigma_max: 80
+  sigma_data: 0.25
+  rho: 7
+  P_mean: -1.2
+  P_std: 1.2
+  S_churn:
+  - 40
+  - 80
+  - 160
+  - 160
+  S_tmin: 0.05
+  S_tmax: 50
+  S_noise: 1.003
+trainer:
+  split_batches: false
+  lr: 0.0005
+  dl_tuple_output_keywords_names:
+  - images
+  - text_embeds
+  - cond_images
+dataset:
+  data_path: /path/to/EchoNet-Dynamic
+  deactivate_cache: false
+  fps: 32
+  duration: 2.0
+  grayscale: false
+dataloader:
+  batch_size: 8
+  num_workers: 8
+wandb:
+  project: EchoDiffusion
+  # entity: your-login
+checkpoint:
+  path: /path/to/checkpoints
+  batch_size: 4
+  cond_scale: 5.0
+  save_every_x_it: 5000

4SCM_v2/merged.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:906f18d1665c69a69f3f65c201d08b63c2557a60ffdac700fec84d2b29c6774e
+size 1523781649

README.md

@@ -1,3 +1,18 @@
 ---
 license: gpl-2.0
 ---
+
+This repository contains 3 models, corresponding to the ones described in *Feature-Conditioned Cascaded Video Diffusion Models for Precise Echocardiogram Synthesis*. Hadrien Reynaud, Mengyun Qiao, Mischa Dombrowski, Thomas Day, Reza Razavi, Alberto Gomez, Paul Leeson and Bernhard Kainz. 2023.
+
+To see all the details, refer to the corresponding github repository: [https://github.com/HReynaud/EchoDiffusion](https://github.com/HReynaud/EchoDiffusion).
+
+The available models are:
+1SCM: Single Stage Cascade Model
+2SCM: Two Stage Cascade Model
+4SCM: Four Stage Cascade Model
+
+All weights files contains the weights of all the diffusion model in the cascade.
+To see a demo of the 1SCM, head to [https://huggingface.co/spaces/HReynaud/echocardiogram-video-diffusion](https://huggingface.co/spaces/HReynaud/echocardiogram-video-diffusion).
+
+In each model folder, you will find:
+- `config.yaml`: the configuration file associated to the model. It contains the hyperparameters of the model.