DocumentAIKit/doc/doc_en/add_new_algorithm_en.md

Add New Algorithm

PaddleOCR decomposes an algorithm into the following parts, and modularizes each part to make it more convenient to develop new algorithms.

• Data loading and processing
• Network
• Post-processing
• Loss
• Metric
• Optimizer

The following will introduce each part separately, and introduce how to add the modules required for the new algorithm.

Data loading and processing

Data loading and processing are composed of different modules, which complete the image reading, data augment and label production. This part is under ppocr/data. The explanation of each file and folder are as follows:

ppocr/data/
├── imaug             # Scripts for image reading, data augment and label production
│   ├── label_ops.py  # Modules that transform the label
│   ├── operators.py  # Modules that transform the image
│   ├──.....
├── __init__.py
├── lmdb_dataset.py   # The dataset that reads the lmdb
└── simple_dataset.py # Read the dataset saved in the form of `image_path\tgt`

PaddleOCR has a large number of built-in image operation related modules. For modules that are not built-in, you can add them through the following steps:

1. Create a new file under the ppocr/data/imaug folder, such as my_module.py.
2. Add code in the my_module.py file, the sample code is as follows:

class MyModule:
    def __init__(self, *args, **kwargs):
        # your init code
        pass

    def __call__(self, data):
        img = data['image']
        label = data['label']
        # your process code

        data['image'] = img
        data['label'] = label
        return data

1. Import the added module in the ppocr/data/imaug/_init_.py file.

All different modules of data processing are executed by sequence, combined and executed in the form of a list in the config file. Such as:

# angle class data process
transforms:
  - DecodeImage: # load image
      img_mode: BGR
      channel_first: False
  - MyModule:
      args1: args1
      args2: args2
  - KeepKeys:
      keep_keys: [ 'image', 'label' ] # dataloader will return list in this order

Network

The network part completes the construction of the network, and PaddleOCR divides the network into four parts, which are under ppocr/modeling. The data entering the network will pass through these four parts in sequence(transforms->backbones-> necks->heads).

├── architectures # Code for building network
├── transforms    # Image Transformation Module
├── backbones     # Feature extraction module
├── necks         # Feature enhancement module
└── heads         # Output module

PaddleOCR has built-in commonly used modules related to algorithms such as DB, EAST, SAST, CRNN and Attention. For modules that do not have built-in, you can add them through the following steps, the four parts are added in the same steps, take backbones as an example:

1. Create a new file under the ppocr/modeling/backbones folder, such as my_backbone.py.
2. Add code in the my_backbone.py file, the sample code is as follows:

import paddle
import paddle.nn as nn
import paddle.nn.functional as F


class MyBackbone(nn.Layer):
    def __init__(self, *args, **kwargs):
        super(MyBackbone, self).__init__()
        # your init code
        self.conv = nn.xxxx

    def forward(self, inputs):
        # your network forward
        y = self.conv(inputs)
        return y

1. Import the added module in the ppocr/modeling/backbones/_init_.py file.

After adding the four-part modules of the network, you only need to configure them in the configuration file to use, such as:

Architecture:
  model_type: rec
  algorithm: CRNN
  Transform:
    name: MyTransform
    args1: args1
    args2: args2
  Backbone:
    name: MyBackbone
    args1: args1
  Neck:
    name: MyNeck
    args1: args1
  Head:
    name: MyHead
    args1: args1

Post-processing

Post-processing realizes decoding network output to obtain text box or recognized text. This part is under ppocr/postprocess. PaddleOCR has built-in post-processing modules related to algorithms such as DB, EAST, SAST, CRNN and Attention. For components that are not built-in, they can be added through the following steps:

1. Create a new file under the ppocr/postprocess folder, such as my_postprocess.py.
2. Add code in the my_postprocess.py file, the sample code is as follows:

import paddle


class MyPostProcess:
    def __init__(self, *args, **kwargs):
        # your init code
        pass

    def __call__(self, preds, label=None, *args, **kwargs):
        if isinstance(preds, paddle.Tensor):
            preds = preds.numpy()
        # you preds decode code
        preds = self.decode_preds(preds)
        if label is None:
            return preds
        # you label decode code
        label = self.decode_label(label)
        return preds, label

    def decode_preds(self, preds):
        # you preds decode code
        pass

    def decode_label(self, preds):
        # you label decode code
        pass

1. Import the added module in the ppocr/postprocess/_init_.py file.

After the post-processing module is added, you only need to configure it in the configuration file to use, such as:

PostProcess:
  name: MyPostProcess
  args1: args1
  args2: args2

Loss

The loss function is used to calculate the distance between the network output and the label. This part is under ppocr/losses. PaddleOCR has built-in loss function modules related to algorithms such as DB, EAST, SAST, CRNN and Attention. For modules that do not have built-in modules, you can add them through the following steps:

1. Create a new file in the ppocr/losses folder, such as my_loss.py.
2. Add code in the my_loss.py file, the sample code is as follows:

import paddle
from paddle import nn


class MyLoss(nn.Layer):
    def __init__(self, **kwargs):
        super(MyLoss, self).__init__()
        # you init code
        pass

    def __call__(self, predicts, batch):
        label = batch[1]
        # your loss code
        loss = self.loss(input=predicts, label=label)
        return {'loss': loss}

1. Import the added module in the ppocr/losses/_init_.py file.

After the loss function module is added, you only need to configure it in the configuration file to use it, such as:

Loss:
  name: MyLoss
  args1: args1
  args2: args2

Metric

Metric is used to calculate the performance of the network on the current batch. This part is under ppocr/metrics. PaddleOCR has built-in evaluation modules related to algorithms such as detection, classification and recognition. For modules that do not have built-in modules, you can add them through the following steps:

1. Create a new file under the ppocr/metrics folder, such as my_metric.py.
2. Add code in the my_metric.py file, the sample code is as follows:

class MyMetric(object):
    def __init__(self, main_indicator='acc', **kwargs):
        # main_indicator is used for select best model
        self.main_indicator = main_indicator
        self.reset()

    def __call__(self, preds, batch, *args, **kwargs):
        # preds is out of postprocess
        # batch is out of dataloader
        labels = batch[1]
        cur_correct_num = 0
        cur_all_num = 0
        # you metric code
        self.correct_num += cur_correct_num
        self.all_num += cur_all_num
        return {'acc': cur_correct_num / cur_all_num, }

    def get_metric(self):
        """
        return metrics {
                 'acc': 0,
                 'norm_edit_dis': 0,
            }
        """
        acc = self.correct_num / self.all_num
        self.reset()
        return {'acc': acc}

    def reset(self):
        # reset metric
        self.correct_num = 0
        self.all_num = 0

1. Import the added module in the ppocr/metrics/_init_.py file.

After the metric module is added, you only need to configure it in the configuration file to use it, such as:

Metric:
  name: MyMetric
  main_indicator: acc

Optimizer

The optimizer is used to train the network. The optimizer also contains network regularization and learning rate decay modules. This part is under ppocr/optimizer. PaddleOCR has built-in Commonly used optimizer modules such as Momentum, Adam and RMSProp, common regularization modules such as Linear, Cosine, Step and Piecewise, and common learning rate decay modules such as L1Decay and L2Decay. Modules without built-in can be added through the following steps, take optimizer as an example:

1. Create your own optimizer in the ppocr/optimizer/optimizer.py file, the sample code is as follows:

from paddle import optimizer as optim


class MyOptim(object):
    def __init__(self, learning_rate=0.001, *args, **kwargs):
        self.learning_rate = learning_rate

    def __call__(self, parameters):
        # It is recommended to wrap the built-in optimizer of paddle
        opt = optim.XXX(
            learning_rate=self.learning_rate,
            parameters=parameters)
        return opt

After the optimizer module is added, you only need to configure it in the configuration file to use, such as:

Optimizer:
  name: MyOptim
  args1: args1
  args2: args2
  lr:
    name: Cosine
    learning_rate: 0.001
  regularizer:
    name: 'L2'
    factor: 0