PointNet++复现

代码运行

分类任务

python train.py

下载数据集并训练,后可接参数,这里使用--batch_size=4

python evaluate.py --num_votes 12 

测试 0.901/0.907

eval mean loss: 0.372334
eval accuracy: 0.901945
eval avg class acc: 0.876547
  airplane:     1.000
   bathtub:     0.920
       bed:     0.950
     bench:     0.800
 bookshelf:     0.930
    bottle:     0.930
      bowl:     1.000
       car:     0.980
     chair:     0.950
      cone:     0.900
       cup:     0.850
   curtain:     0.950
      desk:     0.930
      door:     0.800
   dresser:     0.767
flower_pot:     0.150
 glass_box:     0.920
    guitar:     1.000
  keyboard:     1.000
      lamp:     0.900
    laptop:     1.000
    mantel:     0.960
   monitor:     1.000
night_stand:    0.744
    person:     0.900
     piano:     0.940
     plant:     0.800
     radio:     0.850
range_hood:     0.950
      sink:     0.850
      sofa:     0.970
    stairs:     0.950
     stool:     0.850
     table:     0.670
      tent:     0.950
    toilet:     1.000
  tv_stand:     0.900
      vase:     0.850
  wardrobe:     0.550
      xbox:     0.750

你可能感兴趣的:(tensorflow,深度学习,三维物体识别,pointnet++,三维深度学习,tensorflow)