Add files via upload

inspection.py

@@ -5,6 +5,7 @@ import skimage
 from scipy.stats import skew
 
 filename: str = "example_data_crop"
+use_svd: bool = True
 
 torch_device: torch.device = torch.device(
     "cuda:0" if torch.cuda.is_available() else "cpu"
@@ -18,6 +19,13 @@ print("loading done")
 
 stored_contours = np.load("cells.npy", allow_pickle=True)
 
+if use_svd:
+    data_flat = torch.flatten(
+        data.nan_to_num(nan=0.0).movedim(0, -1),
+        start_dim=0,
+        end_dim=1,
+    )
+
 to_plot = torch.zeros(
     (int(data.shape[0]), int(stored_contours.shape[0])),
     device=torch_device,
@@ -31,9 +39,30 @@ for id in range(0, stored_contours.shape[0]):
         device=torch_device,
         dtype=torch.float32,
     )
+    if use_svd:
+        mask_flat = torch.flatten(
+            mask.unsqueeze(0).nan_to_num(nan=0.0).movedim(0, -1),
+            start_dim=0,
+            end_dim=1,
+        )
+        idx = torch.where(mask_flat > 0)[0]
+        temp = data_flat[idx, :].clone()
+        whiten_mean = torch.mean(temp, dim=-1)
+        temp -= whiten_mean.unsqueeze(-1)
+        svd_u, svd_s, _ = torch.svd_lowrank(temp, q=6)
 
-    ts = (data * mask.unsqueeze(0)).nan_to_num(nan=0.0).sum(dim=(-2, -1)) / mask.sum()
+        whiten_k = (
-    to_plot[:, id] = ts
+            torch.sign(svd_u[0, :]).unsqueeze(0) * svd_u / (svd_s.unsqueeze(0) + 1e-20)
+        )[:, 0]
+
+        temp = temp * whiten_k.unsqueeze(-1)
+        data_svd = temp.movedim(-1, 0).sum(dim=-1)
+        to_plot[:, id] = data_svd
+    else:
+        ts = (data * mask.unsqueeze(0)).nan_to_num(nan=0.0).sum(
+            dim=(-2, -1)
+        ) / mask.sum()
+        to_plot[:, id] = ts
 
 
 skew_value = skew(to_plot.cpu().numpy(), axis=0)