removing unnecessary loop

probreg/filterreg.py

@@ -145,7 +145,7 @@ def __init__(self, source=None, target_normals=None,
 
     @staticmethod
     def _maximization_step(t_source, target, estep_res, trans_p, sigma2, w=0.0,
-                           objective_type='pt2pt', maxiter=10, tol=1.0e-4):
+                           objective_type='pt2pt'):
         m, dim = t_source.shape
         n = target.shape[0]
         assert dim == 2 or dim == 3, "dim must be 2 or 3."
@@ -190,15 +190,14 @@ def __init__(self, source=None, skinning_weight=None,
 
     @staticmethod
     def _maximization_step(t_source, target, estep_res, trans_p, sigma2, w=0.0,
-                           objective_type='', maxiter=50, tol=1.0e-4):
+                           objective_type=''):
         m, dim = t_source.shape
         n6d = dim * 2
         idx_6d = lambda i: slice(i * n6d, (i + 1) * n6d)
         n = target.shape[0]
         n_nodes = trans_p.weights.n_nodes
         assert dim == 3, "dim must be 3."
         m0, m1, m2, _ = estep_res
-        tw = np.zeros(n_nodes * dim * 2)
         c = w / (1.0 - w) * n / m
         m0[m0==0] = np.finfo(np.float32).eps
         m1m0 = np.divide(m1.T, m0).T
@@ -214,29 +213,22 @@ def _maximization_step(t_source, target, estep_res, trans_p, sigma2, w=0.0,
                 jtj_tw += w[0] * w[1] * np.dot(drxdz.T, drxdz)
             a[idx_6d(pair[0]), idx_6d(pair[1])] += jtj_tw
             a[idx_6d(pair[1]), idx_6d(pair[0])] += jtj_tw
-        for _ in range(maxiter):
-            x = np.zeros_like(t_source)
-            for pair in trans_p.weights.pairs_set():
-                for idx in trans_p.weights.in_pair(pair):
-                    w = trans_p.weights[idx]['val']
-                    q0 = dualquat_from_twist(tw[idx_6d(pair[0])])
-                    q1 = dualquat_from_twist(tw[idx_6d(pair[1])])
-                    x[idx] = (w[0] * q0 + w[1] * q1).transform_point(t_source[idx])
+        x = np.zeros_like(t_source)
+        for pair in trans_p.weights.pairs_set():
+            for idx in trans_p.weights.in_pair(pair):
+                x[idx] = t_source[idx]
 
-            rx = np.multiply(drxdx, (x - m1m0).T).T
-            b = np.zeros(n_nodes * n6d)
-            for pair in trans_p.weights.pairs_set():
-                j_tw = np.zeros(n6d)
-                for idx in trans_p.weights.in_pair(pair):
-                    drxdz = drxdx[idx] * dxdz[idx]
-                    w = trans_p.weights[idx]['val']
-                    j_tw += w[0] * np.dot(drxdz.T, rx[idx])
-                b[idx_6d(pair[0])] += j_tw
+        rx = np.multiply(drxdx, (x - m1m0).T).T
+        b = np.zeros(n_nodes * n6d)
+        for pair in trans_p.weights.pairs_set():
+            j_tw = np.zeros(n6d)
+            for idx in trans_p.weights.in_pair(pair):
+                drxdz = drxdx[idx] * dxdz[idx]
+                w = trans_p.weights[idx]['val']
+                j_tw += w[0] * np.dot(drxdz.T, rx[idx])
+            b[idx_6d(pair[0])] += j_tw
 
-            dtw = np.linalg.lstsq(a, b, rcond=None)[0]
-            tw -= dtw
-            if np.linalg.norm(dtw) < tol:
-                break
+        tw = -np.linalg.lstsq(a, b, rcond=None)[0]
 
         dualquats = [dualquat_from_twist(tw[idx_6d(i)]) * dq for i, dq in enumerate(trans_p.dualquats)]
         if not m2 is None: