mindspore/model_zoo/official/cv/crnn

• Contents
• • CRNN Description
  • Model Architecture
  • Dataset
  • • Dataset Prepare
  • 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

• Contents
  • CRNN Description
  • Model Architecture
  • Dataset
    • Dataset Prepare
  • 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

CRNN Description

CRNN was a neural network for image based sequence recognition and its Application to scene text recognition.In this paper, we investigate the problem of scene text recognition, which is among the most important and challenging tasks in image-based sequence recognition. A novel neural network architecture, which integrates feature extraction, sequence modeling and transcription into a unified framework, is proposed. Compared with previous systems for scene text recognition, the proposed architecture possesses four distinctive properties: (1) It is end-to-end trainable, in contrast to most of the existing algorithms whose components are separately trained and tuned. (2) It naturally handles sequences in arbitrary lengths, involving no character segmentation or horizontal scale normalization. (3) It is not confined to any predefined lexicon and achieves remarkable performances in both lexicon-free and lexicon-based scene text recognition tasks. (4) It generates an effective yet much smaller model, which is more practical for real-world application scenarios.

Paper: Baoguang Shi, Xiang Bai, Cong Yao, "An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition", ArXiv, vol. abs/1507.05717, 2015.

Model Architecture

CRNN use a vgg16 structure for feature extraction, the appending with two-layer bidirectional LSTM, finally use CTC to calculate loss. See src/crnn.py for details.

Dataset

Note that you can run the scripts based on the dataset mentioned in original paper or widely used in relevant domain/network architecture. In the following sections, we will introduce how to run the scripts using the related dataset below.

We use five datasets mentioned in the paper.For training, we use the synthetic dataset(MJSynth and SynthText) released by Jaderberg etal as the training data, which contains 8 millions training images and their corresponding ground truth words.For evaluation, we use four popular benchmarks for scene text recognition, nalely ICDAR 2003(IC03),ICDAR2013(IC13),IIIT 5k-word(IIIT5k),and Street View Text(SVT).

Dataset Prepare

For datset IC03, IIIT5k and SVT, the original dataset from the official website can not be used directly in CRNN.

• IC03, the text need to be cropped from the original image according to the words.xml.
• IIIT5k, the annotation need to be extracted from the matlib data file.
• SVT, the text need to be cropped from the original image according to the train.xml or test.xml.

We provide convert_ic03.py, convert_iiit5k.py, convert_svt.py as exmples for the aboving preprocessing which you can refer to.

Environment Requirements

• Hardware（Ascend）
  • Prepare hardware environment with Ascend 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

• 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 [DATASET_NAME] [RANK_TABLE_FILE] [DATASET_PATH]
  
  # evaluation example in Ascend
  $ bash run_eval.sh [DATASET_NAME] [DATASET_PATH] [CHECKPOINT_PATH] [PLATFORM]
  
  # standalone training example in Ascend
  $ bash run_standalone_train.sh [DATASET_NAME] [DATASET_PATH] [PLATFORM]
  

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

  Please follow the instructions in the link below:
  hccl_tools.

Script Description

Script and Sample Code

crnn
├── README.md                                   # Descriptions about CRNN
├── convert_ic03.py                             # Convert the original IC03 daatset
├── convert_iiit5k.py                           # Convert the original IIIT5K dataset
├── convert_svt.py                              # Convert the original SVT dataset
├── requirements.txt                            # Requirements for this dataset
├── scripts
│   ├── run_distribute_train.sh                 # Launch distributed training in Ascend(8 pcs)
│   ├── run_eval.sh                             # Launch evaluation
│   └── run_standalone_train.sh                 # Launch standalone training(1 pcs)
├── src
│   ├── config.py                               # Parameter configuration
│   ├── crnn.py                                 # crnn network definition
│   ├── crnn_for_train.py                       # crnn network with grad, loss and gradient clip
│   ├── dataset.py                              # Data preprocessing for training and evaluation
│   ├── ic03_dataset.py                         # Data preprocessing for IC03
│   ├── ic13_dataset.py                         # Data preprocessing for IC13
│   ├── iiit5k_dataset.py                       # Data preprocessing for IIIT5K
│   ├── loss.py                                 # Ctcloss definition
│   ├── metric.py                               # accuracy metric for crnn network
│   └── svt_dataset.py                          # Data preprocessing for SVT
└── train.py                                    # Training script
├── eval.py                                     # Evaluation Script

Script Parameters

Training Script Parameters

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

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

Parameters Configuration

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

max_text_length": 23,                     # max number of digits in each
"image_width": 100,                        # width of text images
"image_height": 32,                        # height of text images
"batch_size": 64,                            # batch size of input tensor
"epoch_size": 10,                            # only valid for taining, which is always 1
"hidden_size": 256,                          # hidden size in LSTM layers
"learning_rate": 0.02,                       # initial learning rate
"momentum": 0.95,                            # momentum of SGD optimizer
"nesterov": True,                            # enable nesterov in SGD optimizer
"save_checkpoint": True,                     # whether save checkpoint or not
"save_checkpoint_steps": 1000,               # the step interval between two checkpoints.
"keep_checkpoint_max": 30,                   # only keep the last keep_checkpoint_max
"save_checkpoint_path": "./",                # path to save checkpoint
"class_num": 37,                             # dataset class num
"input_size": 512,                           # input size for LSTM layer
"num_step": 24,                              # num step for LSTM layer
"use_dropout": True,                         # whether use dropout
"blank": 36                                  # add blank for classification

Dataset Preparation

• You may refer to "Generate dataset" in Quick Start to automatically generate a dataset, or you may choose to generate a text image 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 CRNN model, only support Ascend now.

bash run_standalone_train.sh [DATASET_NAME] [DATASET_PATH] [PLATFORM](optional)

Distributed Training

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

bash run_distribute_train.sh [DATASET_NAME] [RANK_TABLE_FILE] [DATASET_PATH]

Check the train_parallel0/log.txt and you will get outputs as following:

epoch: 10 step: 14110, loss is 0.0029097411
Epoch time: 2743.688s, per step time: 0.097s

Evaluation Process

Evaluation

• Run run_eval.sh for evaluation.

bash run_eval.sh [DATASET_NAME] [DATASET_PATH] [CHECKPOINT_PATH] [PLATFORM](optional)

Check the eval/log.txt and you will get outputs as following:

result: {'CRNNAccuracy': (0.806)}

Model Description

Performance

Training Performance

Parameters	Ascend 910
Model Version	v1.0
Resource	Ascend 910, CPU 2.60GHz 192cores, Memory 755G
uploaded Date	12/15/2020 (month/day/year)
MindSpore Version	1.0.1
Dataset	Synth
Training Parameters	epoch=10, steps per epoch=14110, batch_size = 64
Optimizer	SGD
Loss Function	CTCLoss
outputs	probability
Loss	0.0029097411
Speed	118ms/step(8pcs)
Total time	557 mins
Parameters (M)	83M (.ckpt file)
Checkpoint for Fine tuning	20.3M (.ckpt file)
Scripts	Link

Evaluation Performance

Parameters	SVT	IIIT5K
Model Version	V1.0	V1.0
Resource	Ascend 910	Ascend 910
Uploaded Date	12/15/2020 (month/day/year)	12/15/2020 (month/day/year)
MindSpore Version	1.0.1	1.0.1
Dataset	SVT	IIIT5K
batch_size	1	1
outputs	ACC	ACC
Accuracy	80.8%	79.7%
Model for inference	83M (.ckpt file)	83M (.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