开源软件名称：

开源软件地址：

开源软件介绍：

MicroOCR

a micro OCR network.

This model can handle complex tasks without lstm, and its accuracy and speed are better than resnet and crnn models.

Task Parameter Reference:

Simple recognition task:nh=32 or 64,depth=2

Chinese recognition task:nh=128 or 256,depth=2

Complex background recognition task:nh=512 or more,depth=2

Model parameters

5000 training pictures, 500 verification pictures.Nh | Depth | Nclass | Params size |Model size(KB)| Total train(epoch) | Word acc:----:|:-----:|:-------:|:---------------:|:------------:|:------------------:|:--------:16 | 2 | 62 | 9.726k | 50 | 99 | 0.78232 | 2 | 62 | 20.414k | 93 | 100 | 0.84264 | 2 | 62 | 44.862k | 190 | 49 | 0.810128 | 2 | 62 | 106.046k | 434 | 45 | 0.882256 | 2 | 62 | 277.566k | 1113 | 50 | 0.872512 | 2 | 62 | 817.214k | 3239 | 45 | 0.8841024 | 2 | 62 | 2.682942M | 10563 | 49 | 0.894

Script Description

MicroOCR├── README.md                                   # descriptions about MicroNet├── simsunb.ttf                                 # font file├── collatefn.py                                # batch data processing├── label_converter.py                          # label converter├── dataset.py                                  # data preprocessing for training and evaluation├── demo.py                                     # inference├── gen_image.py                                # generate image for train and eval├── infer_tool.py                               # inference tool├── logger.py                                   # logger├── keys.py                                     # character├── loss.py                                     # ctcloss definition├── metric.py                                   # accuracy metric for MicroNet├── model.py                                    # MicroNet├── train.py                                    # train the model

Generate data for train and eval

python gen_image.py

Training

python train.py

Inference

python demo.py