supplementary.md

Supplementary information

Classes used in ModelNet40

airplane, bed, bench, bookshelf, bottle, bowl, car,
chair, cone, cup, curtain, door, flower_pot, glass_box,
guitar, keyboard, lamp, laptop, mantel, person, piano,
plant, radio, range_hood, sink, stairs, stool, tent,
toilet, tv_stand, vase, wardrobe, xbox

Train set class data distribution

System used for Training

NVIDIA-GPU Statistics

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 430.40       Driver Version: 430.40       CUDA Version: 10.1     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce RTX 2080    Off  | 00000000:01:00.0  On |                  N/A |
| N/A   40C    P8     9W /  N/A |    185MiB /  7979MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

Computer Statistics:

  Width: 64 bits
  Core
     Description: Motherboard
     Memory
        description: System memory
        size: 62GiB
     CPU
        product: Intel(R) Core(TM) i9-9900K CPU @ 3.60GHz
        vendor: Intel Corp.
        size: 3086MHz
        capacity: 5GHz
        width: 64 bits
     PCI:0
        description: PCI bridge
        product: Xeon E3-1200 v5/E3-1500 v5/6th Gen Core Processor PCIe Controller (x16)
        vendor: Intel Corporation
        version: 0d
        width: 32 bits
        clock: 33MHz
        configuration: driver=pcieport
      DISPLAY
          description: VGA compatible controller
          product: NVIDIA Corporation
          version: a1
          width: 64 bits
          clock: 33MHz
          configuration: driver=nvidia latency=0

Neural Network Training Hyper-parameters

PyTorch and Numpy SEED Value used as 17*19

Neural Network Hyper Parameters used

AutoDecoder Training

-   Epochs: 4
-   Learning Rate: 0.001
-   Batch Size: 32
-   Latent Encoding space size: 256
-   ADAM Optimizer

CompNet Training

-   Epochs: 20
-   Learning Rate: 0.001
-   Batch Size: 16
-   Latent Encoding space size: 256  
-   ADAM Optimizer

CompNet Training (Only for ModelNet40)

-   Epochs: 50
-   Learning Rate: 0.001
-   Batch Size: 16
-   Latent Encoding space size: 256  
-   ADAM Optimizer

Encoding Pair Training

-   Number of iterations: 15
-   Learning Rate: 0.05
-   Batch Size: 16
-   Latent Encoding space size: 256  
-   ADAM Optimizer

Ensemble AutoDecoder CompNet ROC-AUC Results

ROC Curves from the benchmarked classification models

ROC Curves from the Similarity Classification on PointNet7

ROC Curves from the Similarity Classification on PointNet Full

ROC Curves from the Similarity Classification on ModelNet10

ROC Curves from the Similarity Classification on ModelNet40