Contents

• Contents
• PointNet Description
• Model Architecture
• Dataset
• Environment Requirements
• Quick Start
• Script Description
  • Script and Sample Code
  • Script Parameters
  • Training Process
    • Training
  • Evaluation Process
    • Evaluation
    • ONNX Evaluation
  • 310 Inference Process
    • Export MindIR
    • Export ONNX
    • 310 Infer
    • Result
• Model Description
  • Performance
    • Training Performance
    • Inference Performance
• Description of Random Situation
• ModelZoo Homepage

PointNet Description

PointNet was proposed in 2017, it is a hierarchical neural network that applies PointNet recursively on a nested partitioning of the input point set. The author of this paper proposes a method of applying deep learning model directly to point cloud data, which is called pointnet.

Paper: Qi, Charles R., et al. "PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation" arXiv preprint arXiv:1612.00593 (2017).

Model Architecture

For each n × 3 N\times 3N × 3 point cloud input, The network first aligns it spatially through a t-net (rotate to the front), map it to the 64 dimensional space through MLP, align it, and finally map it to the 1024 dimensional space. At this time, there is a 1024 dimensional vector representation for each point, and such vector representation is obviously redundant for a 3-dimensional point cloud. Therefore, the maximum pool operation is introduced at this time to keep only the maximum on all 1024 dimensional channels The one that gets 1 × 1024 1\times 1024 × 1 The vector of 1024 is the global feature of n nn point clouds.

Dataset

Dataset used: Segmentation on A subset of shapenet shapenet

• Data format：txt files
  • Note：Data will be processed in src/dataset.py

Environment Requirements

• Hardware（Ascend）
  • Prepare hardware environment with Ascend processor.
• Framework
  • MindSpore
• For more information, please check the resources below：
  • MindSpore Tutorials
  • MindSpore Python API

Quick Start

After installing MindSpore via the official website, you can start training and evaluation as follows:

# Run stand-alone training
bash scripts/run_standalone_train.sh [DATA_PATH] [CKPT_PATH]
# example:
bash scripts/run_standalone_train.sh '/home/pointnet/shapenetcore_partanno_segmentation_benchmark_v0' '../results'

# Run distributed training
bash scripts/run_distributed_train.sh [RANK_TABLE_FILE] [DATA_PATH] [SAVE_DIR] [PRETRAINDE_CKPT(optional)]
# example:
bash scripts/run_standalone_train.sh hccl_8p_01234567_127.0.0.1.json modelnet40_normal_resampled save pointnet2.ckpt

# Evaluate
bash scripts/run_standalone_eval.sh [DATA_PATH] [MODEL_PATH]
# example:
bash scripts/run_standalone_eval.sh '/home/pointnet/shapenetcore_partanno_segmentation_benchmark_v0' '../results/pointnet_network_epoch_10.ckpt'

Script Description

Script and Sample Code

├── .
    ├── pointnet
        ├── ascend310_infer
        │   ├── inc
        │   │   ├── utils.h
        │   ├── src
        │   │   ├── main.cc
        │   │   ├── utils.cc
        │   ├── build.sh
        ├── scripts
        │   ├── run_distribute_ascend.sh        # launch distributed training with ascend platform (8p)
        │   ├── run_distribute_gpu.sh           # launch distributed training with gpu platform (8p)
        │   ├── run_standalone_eval_ascend.sh   # launch evaluating with ascend platform (1p)
        │   ├── run_standalone_eval_gpu.sh      # launch evaluating with gpu platform (1p)
        │   ├── run_standalone_eval_onnx_gpu.sh # launch evaluating onnx with gpu platform (1p)
        │   ├── run_infer_310.sh                # run 310 infer
        │   ├── run_standalone_train_ascend.sh  # launch standalone training with ascend platform (1p)
        │   └── run_standalone_train_gpu.sh     # launch standalone training with gpu platform (1p)
        ├── src
        │   ├── misc                     # dataset part
        │   ├── dataset.py               # data preprocessing
        │   ├── export.py                # export model
        │   ├── loss.py                  # pointnet loss
        │   ├── network.py               # network definition
        │   └── preprocess.py            # data preprocessing for training
        ├── eval.py                      # eval net
        ├── eval_onnx.py                 # eval onnx
        ├── postprocess.py               # 310 postprocess
        ├── preprocess.py                # 310 preprocess
        ├── README.md
        ├── requirements.txt
        └── train.py                     # train net

Script Parameters

Major parameters in train.py are as follows:
--device_id        # train on which device
--batchSize        # Training batch size.
--nepoch           # Total training epochs.
--learning_rate    # Training learning rate.
--data_url         # The path to the train and evaluation datasets.
--loss_per_epoch   # The times to print loss value per epoch.
--train_url        # The path to save files generated during training.
--model            # The file path to load checkpoint.
--enable_modelarts # Whether to use modelarts.

Training Process

Training

• running on Ascend

  # Run stand-alone training for Ascend
  bash scripts/run_standalone_train_ascend.sh [DATA_PATH] [CKPT_PATH] [DEVICE_ID]
  # example:
  bash scripts/run_standalone_train_ascend.sh ../shapenetcore_partanno_segmentation_benchmark_v0 ./ckpts 1
  
  
  
  # Run distributed training for Ascend
  bash scripts/run_distribution_ascend.sh [RANK_TABLE_FILE] [CKPTS_DIR] [DATA_PATH]
  # example:
  bash scripts/run_distribution_ascend.sh [RANK_TABLE_FILE] ./ckpts ../shapenetcore_partanno_segmentation_benchmark_v0
  
  

• running on GPU

  # Run stand-alone training for GPU
  bash scripts/run_standalone_train_gpu.sh [DATA_PATH] [CKPT_PATH] [DEVICE_ID]
  # example:
  bash scripts/run_standalone_train_gpu.sh ../shapenetcore_partanno_segmentation_benchmark_v0 ./ckpts 1
  # Run distributed training for GPU
  bash scripts/run_distribute_gpu.sh [DATA_PATH] [CKPT_PATH]
  # example:
  bash scripts/run_distribute_gpu.sh ./ckpts ../shapenetcore_partanno_segmentation_benchmark_v0

Distributed training requires the creation of an HCCL configuration file in JSON format in advance. For specific operations, see the instructions in hccl_tools.

After training, the loss value will be achieved as follows:

# train log
Epoch : 1/50  episode : 1/40   Loss : 1.3433  Accuracy : 0.489538 step_time: 1.4269
Epoch : 1/50  episode : 2/40   Loss : 1.2932  Accuracy : 0.541544 step_time: 1.4238
Epoch : 1/50  episode : 3/40   Loss : 1.2558  Accuracy : 0.567900 step_time: 1.4397
Epoch : 1/50  episode : 4/40   Loss : 1.1843  Accuracy : 0.654681 step_time: 1.4235
Epoch : 1/50  episode : 5/40   Loss : 1.1262  Accuracy : 0.726756 step_time: 1.4206
Epoch : 1/50  episode : 6/40   Loss : 1.1000  Accuracy : 0.736225 step_time: 1.4363
Epoch : 1/50  episode : 7/40   Loss : 1.0487  Accuracy : 0.814338 step_time: 1.4457
Epoch : 1/50  episode : 8/40   Loss : 1.0271  Accuracy : 0.782350 step_time: 1.4183
Epoch : 1/50  episode : 9/40   Loss : 0.9777  Accuracy : 0.831025 step_time: 1.4289

...

The model checkpoint will be saved in the 'SAVE_DIR' directory.

Evaluation Process

Evaluation

Before running the command below, please check the checkpoint path used for evaluation.

• running on Ascend

  # Evaluate on ascend
  bash scripts/run_standalone_eval_ascend.sh [DATA_PATH] [MODEL_PATH] [DEVICE_ID]
  # example:
  bash scripts/run_standalone_eval_ascend.sh shapenetcore_partanno_segmentation_benchmark_v0 pointnet.ckpt 1

  You can view the results through the file "log_standalone_eval_ascend". The accuracy of the test dataset will be as follows:

  # grep "mIOU " log_standalone_eval_ascend
  'mIOU for class Chair: 0.869'

• running on GPU

  # Evaluate on GPU
  bash scripts/run_standalone_eval_gpu.sh [DATA_PATH] [MODEL_PATH] [DEVICE_ID]
  # example:
  bash scripts/run_standalone_eval_gpu.sh shapenetcore_partanno_segmentation_benchmark_v0 pointnet.ckpt 1

  You can view the results through the file "log_standalone_eval_gpu". The accuracy of the test dataset will be as follows:

  # grep "mIOU " log_standalone_eval_gpu
  'mIOU for class Chair: 0.869'

ONNX Evaluation

• running on GPU

  # Evaluate on GPU
  bash scripts/run_standalone_eval_onnx_gpu.sh [DATA_PATH] [MODEL_PATH] [DEVICE_ID]
  # example
  bash scripts/run_standalone_eval_onnx_gpu.sh dataset/shapenetcore_partanno_segmentation_benchmark_v0 mindir/pointnet.onnx 1

  You can view the results through the file "log_standalone_eval_onnx_gpu". The accuracy of the test dataset will be as follows:

  # grep "mIOU " log_standalone_eval_onnx_gpu
  'mIOU for class Chair: 0.869'

310 Inference Process

Export MindIR/ONNX

python src/export.py --model [CKPT_PATH] --file_format [FILE_FORMAT]

FILE_FORMAT should be one of ['AIR','MINDIR','ONNX'].

The MindIR/ONNX model will be exported to './mindir/pointnet.mindir'/'./mindir/pointnet.onnx'

310 Infer

before inferring in 310, the mindir model should be exported first. Then run the code below to infer:

bash run_infer_310.sh [MINDIR_PATH] [DATA_PATH] [LABEL_PATH] [DVPP] [DEVICE_ID]
# example:
bash run_infer_310.sh ./mindir/pointnet.mindir ../shapenetcore_partanno_segmentation_benchmark_v0 [LABEL_PATH] N 2

Here, DVPP should be 'N'!

Result

'mIOU : 0.869 '

Model Description

Performance

Training Performance

Parameters	Ascend	GPU(V100(PCIE))
Model Version	PointNet	PointNet
Resource	Ascend 910; CPU 24cores; Memory 256G; OS Euler2.8	NVIDIA RTX Titan-24G
uploaded Date	11/30/2021 (month/day/year)	4/19/2022 (month/day/year)
MindSpore Version	1.3.0	1.3.0 1.5.0 1.6.0
Dataset	A subset of ShapeNet	A subset of ShapeNet
Training Parameters	epoch=50, steps=83, batch_size=64, lr=0.005	epoch=50, steps=83, batch_size=64, lr=0.005
Optimizer	Adam	Adam
Loss Function	NLLLoss	NLLLoss
outputs	probability	probability
Loss	0.01	0.01
Speed	1.5 s/step (1p)	0.19 s/step (1p)
Total time	0.3 h (1p)	10 m (1p)
Checkpoint for Fine tuning	17 MB (.ckpt file)	17 MB (.ckpt file)

Inference Performance

Parameters	Ascend	GPU(V100(PCIE))
Model Version	PointNet	PointNet
Resource	Ascend 910; CPU 24cores; Memory 256G; OS Euler2.8	NVIDIA RTX Titan-24G
Uploaded Date	11/30/2021 (month/day/year)	4/19/2022 (month/day/year)
MindSpore Version	1.3.0	1.3.0 1.5.0 1.6.0
Dataset	A subset of ShapeNet	A subset of ShapeNet
Batch_size	64	64
Outputs	probability	probability
mIOU	86.3% (1p)	86.3% (1p)
Total time	1 min	1 min

Description of Random Situation

We use random seed in train.py

ModelZoo Homepage

Please check the official homepage.