remove excess data

data/brain1_kspace.npy

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:30951c380dffe40b7c851f9631348f783ff6249d98ab17506a7b8f75f6b7286d
-size 419430528

data/brain2_kspace.npy

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:6b1d0cdd6d423aa9b67526803adb9e47263587e3501f379880eb5025aa15336b
-size 524288128

data/knee1_kspace.npy

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:4a303497f5a0e3b1f6398455a2e1fa9b6a92e2699e5739b3b3767778e2ceb7f3
-size 68567168

data/knee2_kspace.npy

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:f7c0492267645e7f0fbd250f533728e91d5979fc626e388846fc48353bfe1793
-size 1028505728

data/prostate2_kspace.npy

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:fc06e928ddd938fefda0e82e295a03b89b14834b32b367bb7e538ffbf79a64ef
-size 1108377728

save_kspace_to_disk.py

@@ -0,0 +1,78 @@
+# hello fellow human, this script is used to save kspace data to disk
+# You may ask why? Well, as it turns out having h5py read the entire .h5 file
+# and then just accessing the kspace data as numpy array takes around 50 seconds for a single file
+# and that's just too slow for me. So I'm going to save the kspace data to disk as numpy arrays
+
+import h5py
+import huggingface_hub as hfh
+import numpy as np
+# datasets
+# osbm/fastmri-prostate 
+# osbm/fastmri-brain
+# osbm/fastmri-knee
+
+# files in the dataset
+# prostate
+# - training_T2_1/file_prostate_AXT2_0002.h5
+# - training_T2_1/file_prostate_AXT2_0015.h5
+# brain
+# - multicoil_train/file_brain_AXFLAIR_200_6002442.h5
+# - multicoil_train/file_brain_AXFLAIR_200_6002487.h5
+# knee
+# - singlecoil_train/file1000015.h5
+# - multicoil_train/file1000015.h5
+
+# Download files
+
+file_paths = {
+    "prostate1": hfh.hf_hub_download(
+        repo_id="osbm/fastmri-prostate",
+        filename="training_T2_1/file_prostate_AXT2_0002.h5",
+        repo_type="dataset",
+        cache_dir="./data"
+    ),
+    "prostate2": hfh.hf_hub_download(
+        repo_id="osbm/fastmri-prostate",
+        filename="training_T2_1/file_prostate_AXT2_0015.h5",
+        repo_type="dataset",
+        cache_dir="./data"
+    ),
+    "brain1": hfh.hf_hub_download(
+        repo_id="osbm/fastmri-brain",
+        filename="multicoil_train/file_brain_AXFLAIR_200_6002442.h5",
+        repo_type="dataset",
+        cache_dir="./data"
+    ),
+    "brain2": hfh.hf_hub_download(
+        repo_id="osbm/fastmri-brain",
+        filename="multicoil_train/file_brain_AXFLAIR_200_6002487.h5",
+        repo_type="dataset",
+        cache_dir="./data"
+    ),
+    "knee1": hfh.hf_hub_download(
+        repo_id="osbm/fastmri-knee",
+        filename="singlecoil_train/file1000015.h5",
+        repo_type="dataset",
+        cache_dir="./data"
+    ),
+    "knee2": hfh.hf_hub_download(
+        repo_id="osbm/fastmri-knee",
+        filename="multicoil_train/file1000015.h5",
+        repo_type="dataset",
+        cache_dir="./data"
+    )
+}
+
+for key, file_path in file_paths.items():
+    print(f"{key}: {file_path}")
+
+    file = h5py.File(file_path, "r")
+    kspace = file["kspace"][()]
+
+    print(kspace.shape)
+    if key.startswith("prostate"):
+        kspace = kspace[0, :, :, :] + kspace[1, :, :, :]
+
+    print(kspace.shape)
+
+    np.save(f"./data/{key}_kspace.npy", kspace)

test.py

@@ -0,0 +1,19 @@
+from fastmri.data.subsample import create_mask_for_mask_type
+from fastmri.data.transforms import apply_mask, to_tensor, center_crop
+import numpy as np
+
+mask_func =create_mask_for_mask_type(
+    mask_type_str="equispaced",
+    center_fractions=[0.37],
+    accelerations=[4]
+)
+
+kspace = np.load("data/prostate1_kspace.npy")
+print(kspace.shape) # (34, 14, 640, 451)
+kspace = to_tensor(kspace)
+print(kspace.shape) # torch.Size([34, 14, 640, 451, 2])
+subsampled_kspace, mask, num_low_frequencies = apply_mask(
+    kspace,
+    mask_func,
+    seed=1
+)