Allow specifying threshold_abs and threshold_rel. This fixes the drop-everything-under-zero bug in ExaSPIM

ffn/inference/seed.py

@@ -316,11 +316,39 @@ def init_coords(self):
 class PolicyMaxPeaks(BaseSeedPolicy):
   """Local peaks of intensity."""
 
+  def __init__(
+      self, canvas, min_distance=3, threshold_abs=0, threshold_rel=0, **kwargs
+  ):
+    """Initialize settings.
+
+    Args:
+      canvas: inference Canvas object.
+      min_distance: forwarded to peak_local_max.
+      threshold_abs: forwarded to peak_local_max.
+      threshold_rel: forwarded to peak_local_max.
+      **kwargs: forwarded to base.
+    """
+    super().__init__(canvas, **kwargs)
+    logging.info(
+        'max peaks: min_distance=%s, threshold_abs=%s, threshold_rel=%s',
+        min_distance,
+        threshold_abs,
+        threshold_rel,
+    )
+    self.min_distance = min_distance
+    self.threshold_abs = threshold_abs
+    self.threshold_rel = threshold_rel
+
   def init_coords(self):
     img = self.canvas.image.astype(np.float32).copy()
     mask = self.get_exclusion_mask()
     img[mask] = 0
-    idxs = _find_peaks(img, min_distance=3, threshold_abs=0, threshold_rel=0)
+    idxs = _find_peaks(
+        img,
+        min_distance=self.min_distance,
+        threshold_abs=self.threshold_abs,
+        threshold_rel=self.threshold_rel,
+    )
     self.coords = np.array(sorted((z, y, x) for z, y, x in idxs))