Model size

[rginpan]

What params inside deeplio, compared with the size of trained model between deeplio and lidar-feat-net, the former one is about 400+MB, and the latter one is only 20+MB, so I am wonder why the size of deeplio model is much bigger than lidar-feat-net model.

deeplio:
dropout: 0.25
pretrained: false
model-path: ""
lidar-feat-net:
name: "lidar-feat-pointseg"
pretrained: false
model-path: ""
requires-grad: true