Hello there, I've run the code for 2d nonrigid point clouds registration, the performance is quite amazing with the default peremeters setting for source and target points. However, when I try to add outliers by setting outlier_ratio=1.1 to source points, as a result, the number of points in each source pointcloud is different, and the algorithm crashes at
write_to_tfrecords({"source": np.asanyarray([i.T for i in a.source_list])[np.random.choice(range(test_num), test_num)], "target": np.asanyarray([i.T for i in a.target_list])}, "temp_test_1.tfrecords")
more excatly, it happens becuase np.asanyarray cannot deal with elements with different shapes.
I read your paper and it's said that outliers are considered and the result seems great, so I'm here to look for some help, how can I modify the codes in order to cope with this problem?