mindspore/model_zoo/official/cv/warpctc

• Contents
• WarpCTC Description
• Model Architecture
• Dataset
• Environment Requirements
• Quick Start
• Script Description
• • Script and Sample Code
  • Script Parameters
  • • Training Script Parameters
    • Parameters Configuration
  • Dataset Preparation
  • Training Process
  • • Training
    • Distributed Training
  • Evaluation Process
  • • Evaluation
• Model Description
• • Performance
  • • Training Performance
    • Evaluation Performance
• Description of Random Situation
• ModelZoo Homepage

Contents

• WarpCTC Description
• Model Architecture
• Dataset
• Environment Requirements
• Quick Start
• Script Description
  • Script and Sample Code
  • Script Parameters
    • Training Script Parameters
    • Parameters Configuration
  • Dataset Preparation
  • Training Process
    • Training
    • Distributed Training
  • Evaluation Process
    • Evaluation
• Model Description
  • Performance
    • Training Performance
    • Evaluation Performance
• Description of Random Situation
• ModelZoo Homepage

WarpCTC Description

This is an example of training WarpCTC with self-generated captcha image dataset in MindSpore.

Model Architecture

WarpCTC is a two-layer stacked LSTM appending with one-layer FC neural network. See src/warpctc.py for details.

Dataset

The dataset is self-generated using a third-party library called captcha, which can randomly generate digits from 0 to 9 in image. In this network, we set the length of digits varying from 1 to 4.

Environment Requirements

• Hardware（Ascend/GPU）
  • Prepare hardware environment with Ascend or GPU processor. If you want to try Ascend, please send the application form to ascend@huawei.com. You will be able to have access to related resources once approved.
• Framework
  • MindSpore
• For more information, please check the resources below：
  • MindSpore Tutorials
  • MindSpore Python API

Quick Start

• Generate dataset.

  Run the script scripts/run_process_data.sh to generate a dataset. By default, the shell script will generate 10000 test images and 50000 train images separately.

   $ cd scripts
   $ sh run_process_data.sh
  
   # after execution, you will find the dataset like the follows:
   .  
   └─warpctc
     └─data
       ├─ train  # train dataset
       └─ test   # evaluate dataset
  

• After the dataset is prepared, you may start running the training or the evaluation scripts as follows:

  • Running on Ascend

  # distribute training example in Ascend
  $ bash run_distribute_train.sh rank_table.json ../data/train
  
  # evaluation example in Ascend
  $ bash run_eval.sh ../data/test warpctc-30-97.ckpt Ascend
  
  # standalone training example in Ascend
  $ bash run_standalone_train.sh ../data/train Ascend
  

  For distributed training, a hccl configuration file with JSON format needs to be created in advance.

  Please follow the instructions in the link below:

  https://gitee.com/mindspore/mindspore/tree/master/model_zoo/utils/hccl_tools.

  • Running on GPU

  # distribute training example in GPU
  $ bash run_distribute_train_for_gpu.sh 8 ../data/train
  
  # standalone training example in GPU
  $ bash run_standalone_train.sh ../data/train GPU
  
  # evaluation example in GPU
  $ bash run_eval.sh ../data/test warpctc-30-97.ckpt GPU
  

Script Description

Script and Sample Code

.
└──warpctc
  ├── README.md
  ├── script
    ├── run_distribute_train.sh         # launch distributed training in Ascend(8 pcs)
    ├── run_distribute_train_for_gpu.sh # launch distributed training in GPU
    ├── run_eval.sh                     # launch evaluation
    ├── run_process_data.sh             # launch dataset generation
    └── run_standalone_train.sh         # launch standalone training(1 pcs)
  ├── src
    ├── config.py                       # parameter configuration
    ├── dataset.py                      # data preprocessing
    ├── loss.py                         # ctcloss definition
    ├── lr_generator.py                 # generate learning rate for each step
    ├── metric.py                       # accuracy metric for warpctc network
    ├── warpctc.py                      # warpctc network definition
    └── warpctc_for_train.py            # warpctc network with grad, loss and gradient clip
  ├── eval.py                           # eval net
  ├── process_data.py                   # dataset generation script
  └── train.py                          # train net

Script Parameters

Training Script Parameters

# distributed training in Ascend
Usage: bash run_distribute_train.sh [RANK_TABLE_FILE] [DATASET_PATH]

# distributed training in GPU
Usage: bash run_distribute_train_for_gpu.sh [RANK_SIZE] [DATASET_PATH]

# standalone training
Usage: bash run_standalone_train.sh [DATASET_PATH] [PLATFORM]

Parameters Configuration

Parameters for both training and evaluation can be set in config.py.

"max_captcha_digits": 4,                    # max number of digits in each 
"captcha_width": 160,                       # width of captcha images
"captcha_height": 64,                       # height of capthca images
"batch_size": 64,                           # batch size of input tensor
"epoch_size": 30,                           # only valid for taining, which is always 1 for inference
"hidden_size": 512,                         # hidden size in LSTM layers
"learning_rate": 0.01,                      # initial learning rate
"momentum": 0.9                             # momentum of SGD optimizer
"save_checkpoint": True,                    # whether save checkpoint or not
"save_checkpoint_steps": 97,                # the step interval between two checkpoints. By default, the last checkpoint will be saved after the last step
"keep_checkpoint_max": 30,                  # only keep the last keep_checkpoint_max checkpoint
"save_checkpoint_path": "./checkpoint",     # path to save checkpoint

Dataset Preparation

• You may refer to "Generate dataset" in Quick Start to automatically generate a dataset, or you may choose to generate a captcha dataset by yourself.

Training Process

• Set options in config.py, including learning rate and other network hyperparameters. Click MindSpore dataset preparation tutorial for more information about dataset.

Training

• Run run_standalone_train.sh for non-distributed training of WarpCTC model, either on Ascend or on GPU.

bash run_standalone_train.sh [DATASET_PATH] [PLATFORM]

Distributed Training

• Run run_distribute_train.sh for distributed training of WarpCTC model on Ascend.

bash run_distribute_train.sh [RANK_TABLE_FILE] [DATASET_PATH]

• Run run_distribute_train_gpu.sh for distributed training of WarpCTC model on GPU.

bash run_distribute_train_gpu.sh [RANK_SIZE] [DATASET_PATH]

Evaluation Process

Evaluation

• Run run_eval.sh for evaluation.

bash run_eval.sh [DATASET_PATH] [CHECKPOINT_PATH] [PLATFORM]

Model Description

Performance

Training Performance

Parameters	Ascend 910	GPU
Model Version	v1.0	v1.0
Resource	Ascend 910，CPU 2.60GHz 56cores，Memory 314G	GPU(Tesla V100 SXM2)，CPU 2.1GHz 24cores，Memory 128G /
uploaded Date	07/01/2020 (month/day/year)	08/01/2020 (month/day/year)
MindSpore Version	0.5.0-alpha	0.6.0-alpha
Dataset	Captcha	Captcha
Training Parameters	epoch=30, steps per epoch=98, batch_size = 64	epoch=30, steps per epoch=98, batch_size = 64
Optimizer	SGD	SGD
Loss Function	CTCLoss	CTCLoss
outputs	probability	probability
Loss	0.0000157	0.0000246
Speed	980ms/step（8pcs）	150ms/step（8pcs）
Total time	30 mins	5 mins
Parameters (M)	2.75	2.75
Checkpoint for Fine tuning	20.3M (.ckpt file)	20.3M (.ckpt file)
Scripts	Link	Link

Evaluation Performance

Parameters	WarpCTC
Model Version	V1.0
Resource	Ascend 910
Uploaded Date	08/01/2020 (month/day/year)
MindSpore Version	0.6.0-alpha
Dataset	Captcha
batch_size	64
outputs	ACC
Accuracy	99.0%
Model for inference	20.3M (.ckpt file)

Description of Random Situation

In dataset.py, we set the seed inside “create_dataset" function. We also use random seed in train.py for weight initialization.

ModelZoo Homepage

Please check the official homepage.