the current version sometimes ends up with pointmatches having a high…

…er error than epsilon if one re-fits the model using the inliers. It has to iterate until the number of inliers didn't change once

mpicbg/src/main/java/mpicbg/models/AbstractModel.java

@@ -323,7 +323,7 @@ final public < P extends PointMatch >boolean ransac(
 
 			int numInliers = 0;
 			boolean isGood = m.test( candidates, tempInliers, epsilon, minInlierRatio );
-			while ( isGood && numInliers < tempInliers.size() )
+			while ( isGood && numInliers != tempInliers.size() )
 			{
 				numInliers = tempInliers.size();
 				try { m.fit( tempInliers ); }