retrurn idx in nms; Add scores (#19)

spotiflow/lib/point_nms.cpp

@@ -145,7 +145,7 @@ static PyObject *c_point_nms_2d(PyObject *self, PyObject *args)
             const long k = ret_matches[j].first;
             const float dist = ret_matches[j].second;
             if ((k != i)  && (dist < min_distance_squared)) {
-                // std::cout << "suppressed: " << k << " "<< *(float *)PyArray_GETPTR2(points, k, 0) << " (y,x) = " << *(float *)PyArray_GETPTR2(points, k, 0) << " distance " << dist << std::endl;
+                // std::cout << "suppressed: " << k << " "<< *(float *)PyArray_GETPTR2(points, k, 0) << "  " << *(float *)PyArray_GETPTR2(points, k, 1) << " distance " << dist << std::endl;
                 suppressed[k] = true;
             }
         }

spotiflow/utils/peaks.py

@@ -60,13 +60,14 @@ def nms_points_2d(
     if not scores.ndim == 1:
         raise ValueError("scores must be a array of shape (N,)")
 
-    idx = np.argsort(scores, kind="stable")
+    idx = np.argsort(scores, kind="stable")[::-1]
     points = points[idx]
     scores = scores[idx]
-
+    
     points = np.ascontiguousarray(points, dtype=np.float32)
     inds = c_point_nms_2d(points, np.float32(min_distance))
-    return points[inds].copy()
+    inds = idx[inds]
+    return inds
 
 def nms_points_3d(
     points: np.ndarray, scores: np.ndarray = None, min_distance: int = 2
@@ -100,13 +101,14 @@ def nms_points_3d(
     if not scores.ndim == 1:
         raise ValueError("scores must be a array of shape (N,)")
 
-    idx = np.argsort(scores, kind="stable")
+    idx = np.argsort(scores, kind="stable")[::-1]
     points = points[idx]
     scores = scores[idx]
 
     points = np.ascontiguousarray(points, dtype=np.float32)
     inds = c_point_nms_3d(points, np.float32(min_distance))
-    return points[inds].copy()
+    inds = idx[inds]
+    return inds
 
 def maximum_filter_2d(image: np.ndarray, kernel_size: int = 3) -> np.ndarray:
 
@@ -398,6 +400,7 @@ def local_peaks(
     exclude_border=True,
     threshold_abs=None,
     threshold_rel=None,
+    use_score:bool=False
 ):
     if not image.ndim in [2, 3]:
         raise ValueError("Image must be 2D")
@@ -427,8 +430,14 @@ def local_peaks(
     coord = np.nonzero(mask)
     coord = np.stack(coord, axis=1)
 
-    points = nms_fun(coord, min_distance=min_distance)
-    return points
+    if use_score:
+        scores = image[mask] if mask.sum() > 0 else None
+    else:
+        scores = None
+
+    idx = nms_fun(coord, scores=scores, min_distance=min_distance)
+    coord = coord[idx].copy()
+    return coord