Add universal + breast vSHARP checkpoints; refresh READMEs and inference YAMLs

README.md

@@ -12,27 +12,29 @@ pipeline_tag: image-to-image
 
 # DIRECT — vSHARP multi-anatomy
 
-Pretrained [vSHARP](https://arxiv.org/abs/2309.09954) (variable Splitting Half-quadratic ADMM algorithm for Reconstruction of inverse Problems) models for multi-coil MRI reconstruction across four anatomies. Each release is an inference-ready pair:
+Pretrained [vSHARP](https://arxiv.org/abs/2309.09954) models for multi-coil MRI reconstruction. Each release is an inference-ready pair:
 
 ```text
-vsharp_<anatomy>.yaml   # inference-only DIRECT config
-vsharp_<anatomy>.pt     # weights (ModConv layout)
+vsharp_<name>.yaml   # inference-only DIRECT config
+vsharp_<name>.pt     # weights (ModConv layout)
 ```
 
 Paper: [arXiv:2309.09954](https://arxiv.org/abs/2309.09954) · Framework: [DIRECT](https://github.com/NKI-AI/direct)
 
 ## Datasets
 
-| Model | Dataset | Link |
-|-------|---------|------|
+| Model | Training data | Link |
+|-------|---------------|------|
 | `vsharp_brain` | fastMRI brain (multi-coil) | [fastmri.med.nyu.edu](https://fastmri.med.nyu.edu/) |
 | `vsharp_knee` | fastMRI knee (multi-coil) | [fastmri.med.nyu.edu](https://fastmri.med.nyu.edu/) |
 | `vsharp_prostate` | fastMRI prostate (multi-coil) | [fastmri.med.nyu.edu](https://fastmri.med.nyu.edu/) |
+| `vsharp_breast` | fastMRI breast (multi-coil) | [fastmri.med.nyu.edu](https://fastmri.med.nyu.edu/) |
 | `vsharp_cardiac` | CMRxRecon 2023 cine (multi-coil) | [cmrxrecon.github.io](https://cmrxrecon.github.io/) |
+| `vsharp_universal` | Mixed: fastMRI brain/knee/prostate/breast + CMRxRecon 2023/2024/2025 | [fastMRI](https://fastmri.med.nyu.edu/) · [CMRxRecon](https://cmrxrecon.github.io/) |
 
 ## Training protocol
 
-These models were **not** trained at a single fixed acceleration or scheme. Every anatomy used the same mixed schedule:
+These models were **not** trained at a single fixed acceleration or scheme. All used the same mixed schedule:
 
 | | Values |
 |---|---|
@@ -40,6 +42,8 @@ These models were **not** trained at a single fixed acceleration or scheme. Ever
 | Center fractions (ACS) | `0.16, 0.08, 0.06, 0.04, 0.02` (paired with \(R\): 2→0.16, 4→0.08, 6→0.06, 8→0.04, 10→0.02) |
 | Sampling schemes | FastMRIEquispaced, FastMRIRandom, Gaussian1D, Gaussian2D, VariableDensityPoisson, Radial |
 
+`vsharp_universal` was trained jointly across the anatomies / challenges above; anatomy-specific models were trained on one dataset each.
+
 ## Files & default inference masks
 
 Released YAMLs are **inference-only** (no `training` / `validation` blocks). Each pins **one** acceleration and **one** ACS fraction — DIRECT’s mask sampler draws randomly from lists, so multi-\(R\) lists belong in training only.
@@ -49,7 +53,9 @@ Released YAMLs are **inference-only** (no `training` / `validation` blocks). Eac
 | `vsharp_brain.{yaml,pt}` | FastMRIRandom | 4× / 0.08 |
 | `vsharp_knee.{yaml,pt}` | FastMRIEquispaced | 4× / 0.08 |
 | `vsharp_prostate.{yaml,pt}` | FastMRIEquispaced | 4× / 0.08 |
+| `vsharp_breast.{yaml,pt}` | Radial | 4× / 0.08 |
 | `vsharp_cardiac.{yaml,pt}` | FastMRIEquispaced | 4× / 0.08 |
+| `vsharp_universal.{yaml,pt}` | FastMRIEquispaced | 4× / 0.08 |
 
 ## Changing acceleration or scheme
 
@@ -122,7 +128,7 @@ If you use these models or [DIRECT](https://github.com/NKI-AI/direct), please ci
 
 ```bibtex
 @article{yiasemis2024vsharp,
-  title={vSHARP: variable Splitting Half-quadratic ADMM algorithm for Reconstruction of inverse Problems},
+  title={vSHARP: variable Splitting Half-quadratic {ADMM} algorithm for Reconstruction of inverse Problems},
   author={Yiasemis, George and Moriakov, Nikita and S{\'a}nchez, Clara I. and Sonke, Jan-Jakob and Teuwen, Jonas},
   journal={Magnetic Resonance Materials in Physics, Biology and Medicine},
   year={2024},

vsharp_brain.yaml

@@ -23,6 +23,7 @@ model:
     image_unet_num_filters: 16
     image_unet_num_pool_layers: 5
     image_unet_dropout: 0.0
+    image_unet_conv_out_bias: true
 additional_models:
     sensitivity_model:
         model_name: unet.unet_2d.UnetModel2d
@@ -31,6 +32,7 @@ additional_models:
         num_filters: 16
         num_pool_layers: 5
         dropout_probability: 0.0
+        conv_out_bias: true
 physics:
     forward_operator: fft2
     backward_operator: ifft2

vsharp_breast.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ea6d1e8bc7f3f6c4d1c22415a26dd7d41378039287663cd83237ea3dba0713a4
+size 529175381

vsharp_breast.yaml

@@ -0,0 +1,59 @@
+model:
+    model_name: vsharp.vsharp.VSharpNet
+    num_steps: 16
+    num_steps_dc_gd: 12
+    image_init: SENSE
+    no_parameter_sharing: true
+    auxiliary_steps: -1
+    image_model_architecture: UNET
+    initializer_channels:
+        - 32
+        - 32
+        - 32
+        - 64
+        - 64
+    initializer_dilations:
+        - 1
+        - 1
+        - 2
+        - 4
+        - 8
+    initializer_multiscale: 1
+    initializer_activation: PRELU
+    image_unet_num_filters: 16
+    image_unet_num_pool_layers: 5
+    image_unet_dropout: 0.0
+    image_unet_conv_out_bias: true
+additional_models:
+    sensitivity_model:
+        model_name: unet.unet_2d.UnetModel2d
+        in_channels: 2
+        out_channels: 2
+        num_filters: 16
+        num_pool_layers: 5
+        dropout_probability: 0.0
+        conv_out_bias: true
+physics:
+    forward_operator: fft2
+    backward_operator: ifft2
+inference:
+    batch_size: 1
+    crop: header
+    dataset:
+        name: FastMRI
+        transforms:
+            use_seed: true
+            delete_kspace: false
+            masking:
+                name: Radial
+                accelerations:
+                    - 4
+                center_fractions:
+                    - 0.08
+            cropping:
+                image_center_crop: false
+            sensitivity_map_estimation:
+                estimate_sensitivity_maps: true
+            normalization:
+                scaling_key: masked_kspace
+                scale_percentile: 0.995

vsharp_cardiac.yaml

@@ -23,6 +23,7 @@ model:
     image_unet_num_filters: 16
     image_unet_num_pool_layers: 5
     image_unet_dropout: 0.0
+    image_unet_conv_out_bias: true
 additional_models:
     sensitivity_model:
         model_name: unet.unet_2d.UnetModel2d
@@ -31,6 +32,7 @@ additional_models:
         num_filters: 16
         num_pool_layers: 5
         dropout_probability: 0.0
+        conv_out_bias: true
 physics:
     forward_operator: fft2
     backward_operator: ifft2

vsharp_knee.yaml

@@ -23,6 +23,7 @@ model:
     image_unet_num_filters: 16
     image_unet_num_pool_layers: 5
     image_unet_dropout: 0.0
+    image_unet_conv_out_bias: true
 additional_models:
     sensitivity_model:
         model_name: unet.unet_2d.UnetModel2d
@@ -31,6 +32,7 @@ additional_models:
         num_filters: 16
         num_pool_layers: 5
         dropout_probability: 0.0
+        conv_out_bias: true
 physics:
     forward_operator: fft2
     backward_operator: ifft2

vsharp_prostate.yaml

@@ -23,6 +23,7 @@ model:
     image_unet_num_filters: 16
     image_unet_num_pool_layers: 5
     image_unet_dropout: 0.0
+    image_unet_conv_out_bias: true
 additional_models:
     sensitivity_model:
         model_name: unet.unet_2d.UnetModel2d
@@ -31,6 +32,7 @@ additional_models:
         num_filters: 16
         num_pool_layers: 5
         dropout_probability: 0.0
+        conv_out_bias: true
 physics:
     forward_operator: fft2
     backward_operator: ifft2

vsharp_universal.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28dcddf75a16f12094c97e45e7b5e43d81b4b69bc64d6273e1e330a91c9c0f60
+size 529176923

vsharp_universal.yaml

@@ -0,0 +1,59 @@
+model:
+    model_name: vsharp.vsharp.VSharpNet
+    num_steps: 16
+    num_steps_dc_gd: 12
+    image_init: SENSE
+    no_parameter_sharing: true
+    auxiliary_steps: -1
+    image_model_architecture: UNET
+    initializer_channels:
+        - 32
+        - 32
+        - 32
+        - 64
+        - 64
+    initializer_dilations:
+        - 1
+        - 1
+        - 2
+        - 4
+        - 8
+    initializer_multiscale: 1
+    initializer_activation: PRELU
+    image_unet_num_filters: 16
+    image_unet_num_pool_layers: 5
+    image_unet_dropout: 0.0
+    image_unet_conv_out_bias: true
+additional_models:
+    sensitivity_model:
+        model_name: unet.unet_2d.UnetModel2d
+        in_channels: 2
+        out_channels: 2
+        num_filters: 16
+        num_pool_layers: 5
+        dropout_probability: 0.0
+        conv_out_bias: true
+physics:
+    forward_operator: fft2
+    backward_operator: ifft2
+inference:
+    batch_size: 1
+    crop: header
+    dataset:
+        name: FastMRI
+        transforms:
+            use_seed: true
+            delete_kspace: false
+            masking:
+                name: FastMRIEquispaced
+                accelerations:
+                    - 4
+                center_fractions:
+                    - 0.08
+            cropping:
+                image_center_crop: false
+            sensitivity_map_estimation:
+                estimate_sensitivity_maps: true
+            normalization:
+                scaling_key: masked_kspace
+                scale_percentile: 0.995