is there any bug with the pointsift_model use in the code?

[laiming997]

According the https://blog.csdn.net/JiangZuning/article/details/87904417, the point [B, C,64] upsample to the [B, C, 256], the code has used three times of pointsift_model, like that:

l2_points = pointnet_fp_module(l2_xyz, l3_xyz, l2_points, l3_points, [512,512], is_training, bn_decay, scope='fa_layer2')
   _, l2_points_1, _ = pointSIFT_module(l2_xyz, l2_points, radius=0.5, out_channel=512, is_training=is_training, bn_decay=bn_decay, scope='fa_layer2_c0')
    _, l2_points_2, _ = pointSIFT_module(l2_xyz, l2_points, radius=0.5, out_channel=512, is_training=is_training, bn_decay=bn_decay, scope='fa_layer2_c1')
    _, l2_points_3, _ = pointSIFT_module(l2_xyz, l2_points, radius=0.5, out_channel=512, is_training=is_training, bn_decay=bn_decay, scope='fa_layer2_c2')

but the parameter is the same ,and the second input is also l2_points instead of the first pointSIFT_module output, the third iput is l2_points of instead of l2_points_2. This can't enlarge the receptive field, l2_points_1、l2_points_2、l2_points_3 has the same receptive field.

and the same procedure in the point [B, C,256] upsample to the [B, C, 1024],the code is :

  l1_points = pointnet_fp_module(l1_xyz, l2_xyz, l1_points, l2_points, [256,256], is_training, bn_decay, scope='fa_layer3')
    _, l1_points_1, _ = pointSIFT_module(l1_xyz, l1_points, radius=0.25, out_channel=256, is_training=is_training, bn_decay=bn_decay, scope='fa_layer3_c0')
    _, l1_points_2, _ = pointSIFT_module(l1_xyz, l1_points_1, radius=0.25, out_channel=256, is_training=is_training, bn_decay=bn_decay, scope='fa_layer3_c1')
    l1_points = tf.concat([l1_points_1, l1_points_2], axis=-1)

the second pointSIFT_module input is the first pointSIFT_module output l1_points_1, so it can enlarge the receptive field.
so I think is that a bug of the code?or is just a trick for the training.

[sjtuytc]

It's a trick for the training.