added model use

README.md

@@ -20,9 +20,56 @@ This model can be used to reconstruct brain data from Siemens scanner at acceler
 It was shown [here](https://arxiv.org/abs/2106.00753), that it can generalize well, although further tests are required.
 
 ## How to use
-This model can be loaded using the following repo: https://github.com/zaccharieramzi/fastmri-reproducible-benchmark
+This model can be loaded using the following repo: https://github.com/zaccharieramzi/fastmri-reproducible-benchmark.
+After cloning the repo, `git clone https://github.com/zaccharieramzi/fastmri-reproducible-benchmark`, you can install the package via `pip install fastmri-reproducible-benchmark`.
 The framework is TensorFlow.
-Example notebooks and detailed use guides are WIP.
+
+You can initialize and load the model weights as follows:
+```python
+import tensorflow as tf
+
+from fastmri_recon.models.subclassed_models.denoisers.proposed_params import get_model_specs
+from fastmri_recon.models.subclassed_models.xpdnet import XPDNet
+
+
+n_primal = 5
+model_fun, model_kwargs, n_scales, res = [
+        (model_fun, kwargs, n_scales, res)
+        for m_name, m_size, model_fun, kwargs, _, n_scales, res in get_model_specs(n_primal=n_primal, force_res=False)
+        if m_name == 'MWCNN' and m_size == 'medium'
+][0]
+model_kwargs['use_bias'] = False
+run_params = dict(
+    n_primal=n_primal,
+    multicoil=True,
+    n_scales=n_scales,
+    refine_smaps=True,
+    refine_big=True,
+    res=res,
+    output_shape_spec=True,
+    n_iter=25,
+)
+model = XPDNet(model_fun, model_kwargs, **run_params)
+kspace_size = [1, 1, 320, 320]
+inputs = [
+    tf.zeros(kspace_size + [1], dtype=tf.complex64),  # kspace
+    tf.zeros(kspace_size, dtype=tf.complex64),  # mask
+    tf.zeros(kspace_size, dtype=tf.complex64),  # smaps
+    tf.constant([[320, 320]]),  # shape
+]
+model(inputs)
+model.load_weights('xpdnet_sense_brain__af4_i25_compound_mssim_rf_smb_MWCNNmedium_1601987069-100.h5')
+```
+
+Using the model is then as simple as:
+```python
+model([
+    kspace,  # shape: [n_slices, n_coils, n_rows, n_cols, 1]
+    mask,  # shape: [n_slices, n_coils, n_rows, n_cols]
+    smaps,  # shape: [n_slices, n_coils, n_rows, n_cols]
+    shape,  # shape: [n_slices, 2]
+])
+```
 
 ## Limitations and bias
 The limitations and bias of this model have not been properly investigated.