### Install Dependencies

Source: https://github.com/jaidedai/easyocr/blob/master/trainer/craft/README.md

Installs all necessary packages for training CRAFT models. Ensure you have a requirements.txt file in your current directory.

```bash
pip install -r requirements.txt
```

--------------------------------

### Install EasyOCR (Development)

Source: https://github.com/jaidedai/easyocr/blob/master/README.md

Install the latest development release of EasyOCR directly from its GitHub repository.

```bash
pip install git+https://github.com/JaidedAI/EasyOCR.git
```

--------------------------------

### Install EasyOCR (Stable)

Source: https://github.com/jaidedai/easyocr/blob/master/README.md

Install the latest stable release of EasyOCR using pip.

```bash
pip install easyocr
```

--------------------------------

### Get EasyOCR Installation Path

Source: https://github.com/jaidedai/easyocr/blob/master/easyocr/DBNet/README.md

Determine the installation directory of the EasyOCR library in Python.

```python
> import os
> import easyocr
> print(os.dirname(easyocr.__file__))
```

--------------------------------

### Initialize Training with Configuration

Source: https://github.com/jaidedai/easyocr/blob/master/trainer/trainer.ipynb

Loads the specified configuration file and starts the EasyOCR training process with automatic mixed precision disabled.

```python
opt = get_config("config_files/en_filtered_config.yaml")
train(opt, amp=False)
```

--------------------------------

### EasyOCR Command-Line Interface Examples

Source: https://context7.com/jaidedai/easyocr/llms.txt

Provides examples of using the EasyOCR command-line interface for various OCR tasks, including language selection, detail modes, and specific configurations.

```bash
# Basic usage: English + Simplified Chinese, detail mode
easyocr -l en ch_sim -f street_sign.jpg --detail=1 --gpu=True
```

```bash
# CPU-only, simple output (text only)
easyocr -l en -f receipt.jpg --detail=0 --gpu=False
```

```bash
# Beam search decoder, batch size 4
easyocr -l en de fr -f document.jpg --decoder=beamsearch --beamWidth=5 --batch_size=4
```

```bash
# Restrict to digits only
easyocr -l en -f invoice.jpg --allowlist=0123456789
```

```bash
# Paragraph mode with output format JSON
easyocr -l en -f article.jpg --paragraph=True --output_format=json
```

```bash
# DBNet18 detector, custom model directory
easyocr \
  -l en \
  -f photo.jpg \
  --detect_network=dbnet18 \
  --model_storage_directory=/opt/models \
  --recog_network=standard \
  --gpu=True \
  --detail=1
```

--------------------------------

### Check Conda Installation

Source: https://github.com/jaidedai/easyocr/blob/master/easyocr/DBNet/README.md

Verify if conda is installed on your system by checking its version.

```bash
> conda --version
```

--------------------------------

### Train CRAFT with Custom Dataset (Multi-GPU)

Source: https://github.com/jaidedai/easyocr/blob/master/trainer/craft/README.md

Starts training the CRAFT model using a custom dataset or a combination of SynthText and IC15. This command is for multi-GPU usage.

```bash
CUDA_VISIBLE_DEVICES=0,1 python3 train_distributed.py --yaml=custom_data_train
```

--------------------------------

### Install GCC Compiler (Debian/Ubuntu)

Source: https://github.com/jaidedai/easyocr/blob/master/easyocr/DBNet/README.md

Install the build-essential package, which includes GCC, on Debian/Ubuntu-based systems. This is a prerequisite for compiling DCN operators.

```bash
> apt-get install build-essential
```

```bash
> sudo apt-get install build-essential
```

--------------------------------

### Train CRAFT with Custom Dataset (Single GPU)

Source: https://github.com/jaidedai/easyocr/blob/master/trainer/craft/README.md

Starts training the CRAFT model using a custom dataset or a combination of SynthText and IC15. This command is for single GPU usage.

```bash
CUDA_VISIBLE_DEVICES=0 python3 train.py --yaml=custom_data_train
```

--------------------------------

### Verify Docker Prerequisites

Source: https://github.com/jaidedai/easyocr/blob/master/easyocr/DBNet/README.md

Check if GCC and NVCC (for CUDA) are installed within a Docker image.

```bash
gcc --version
```

```bash
nvcc --version
```

--------------------------------

### Install NVCC and CUDA Toolkit with Conda

Source: https://github.com/jaidedai/easyocr/blob/master/easyocr/DBNet/README.md

Install NVCC and the CUDA developer toolkit using the conda package manager. This is an alternative method for setting up CUDA prerequisites.

```bash
> conda install -c conda-forge cudatoolkit-dev
```

--------------------------------

### Check GCC Version

Source: https://github.com/jaidedai/easyocr/blob/master/easyocr/DBNet/README.md

Verify if GCC compiler is installed on your system. This is a prerequisite for compiling DCN operators.

```bash
> gcc --version
```

--------------------------------

### Navigate to DCN Directory

Source: https://github.com/jaidedai/easyocr/blob/master/easyocr/DBNet/README.md

Change the current directory to the DCN operator source files within the EasyOCR installation.

```bash
> cd /home/username/anaconda3/lib/python3.8/site-packages/easyocr/DBNet/assets/ops/dcn
```

--------------------------------

### OCR Output Format

Source: https://github.com/jaidedai/easyocr/blob/master/README.md

Example of the detailed output format from EasyOCR, including bounding box coordinates, detected text, and confidence level.

```bash
[([[189, 75], [469, 75], [469, 165], [189, 165]], '愚园路', 0.3754989504814148),
([[86, 80], [134, 80], [134, 128], [86, 128]], '西', 0.40452659130096436),
([[517, 81], [565, 81], [565, 123], [517, 123]], '东', 0.9989598989486694),
([[78, 126], [136, 126], [136, 156], [78, 156]], '315', 0.8125889301300049),
([[514, 126], [574, 126], [574, 156], [514, 156]], '309', 0.4971577227115631),
([[226, 170], [414, 170], [414, 220], [226, 220]], 'Yuyuan Rd.', 0.8261902332305908),
([[79, 173], [125, 173], [125, 213], [79, 213]], 'W', 0.9848111271858215),
([[529, 173], [569, 173], [569, 213], [529, 213]], 'E', 0.8405593633651733)]
```

--------------------------------

### Run EasyOCR Unit Tests with Python Script

Source: https://github.com/jaidedai/easyocr/blob/master/unit_test/README.md

Execute unit tests for EasyOCR using the provided Python script. Specify the EasyOCR package directory, verbosity level, test data path, and example data directory as arguments.

```bash
python ./unit_test/run_unit_test.py --easyocr ./easyocr --verbose 2 --test ./unit_test/EasyOcrUnitTestPackage.pickle --data_dir ./examples 
```

--------------------------------

### Train CRAFT with SynthText Dataset

Source: https://github.com/jaidedai/easyocr/blob/master/trainer/craft/README.md

Initiates training for the CRAFT model using the SynthText dataset from scratch. This command is for single GPU usage.

```bash
CUDA_VISIBLE_DEVICES=0 python3 trainSynth.py --yaml=syn_train
```

--------------------------------

### Initialize Reader with DBNet Detector

Source: https://github.com/jaidedai/easyocr/blob/master/releasenotes.md

Use this to initialize the EasyOCR Reader with the DBNet text detector. Currently requires GPU.

```python
reader = easyocr.Reader(['en'], detect_network = 'dbnet18')
```

--------------------------------

### Switch Detection Network at Runtime

Source: https://context7.com/jaidedai/easyocr/llms.txt

Demonstrates how to switch the detection network (e.g., from 'craft' to 'dbnet18') after a Reader object has been initialized, without needing to re-create the object.

```python
import easyocr

reader = easyocr.Reader(['en'], detect_network='craft')

# Run with CRAFT detector
results_craft = reader.readtext('image.jpg')

# Switch to DBNet18 detector
reader.setDetector('dbnet18')

# Run with DBNet18 detector
results_db = reader.readtext('image.jpg')
```

--------------------------------

### Simplified OCR Result

Source: https://github.com/jaidedai/easyocr/blob/master/README.md

Example of the simplified output format from EasyOCR when `detail=0` is used.

```bash
['愚园路', '西', '东', '315', '309', 'Yuyuan Rd.', 'W', 'E']
```

--------------------------------

### Configure test parameters

Source: https://github.com/jaidedai/easyocr/blob/master/unit_test/demo.ipynb

Sets up paths and variables for the unit test configuration.

```python
easyocr_module = "../easyocr"
verbose = 2
test_data = "./data/EasyOcrUnitTestPackage.pickle"
image_data_dir = "../examples"
```

--------------------------------

### Load and Process Configuration from YAML

Source: https://github.com/jaidedai/easyocr/blob/master/trainer/trainer.ipynb

Loads training configuration from a YAML file, processes custom character sets if specified, and creates the experiment directory.

```python
def get_config(file_path):
    with open(file_path, 'r', encoding="utf8") as stream:
        opt = yaml.safe_load(stream)
    opt = AttrDict(opt)
    if opt.lang_char == 'None':
        characters = ''
        for data in opt['select_data'].split('-'):
            csv_path = os.path.join(opt['train_data'], data, 'labels.csv')
            df = pd.read_csv(csv_path, sep='^([^,]+),', engine='python', usecols=['filename', 'words'], keep_default_na=False)
            all_char = ''.join(df['words'])
            characters += ''.join(set(all_char))
        characters = sorted(set(characters))
        opt.character= ''.join(characters)
    else:
        opt.character = opt.number + opt.symbol + opt.lang_char
    os.makedirs(f'./saved_models/{opt.experiment_name}', exist_ok=True)
    return opt
```

--------------------------------

### Simplified OCR Output

Source: https://github.com/jaidedai/easyocr/blob/master/README.md

Perform OCR with `detail=0` to get a simplified output containing only the detected text strings.

```python
reader.readtext('chinese.jpg', detail = 0)
```

--------------------------------

### Initialize EasyOCR Reader with Custom Models

Source: https://github.com/jaidedai/easyocr/blob/master/README.md

Instantiate the EasyOCR Reader by specifying the desired languages and optionally custom detection and recognition models. This allows for flexible configuration of OCR pipelines.

```python
reader = easyocr.Reader(['en'], detection='DB', recognition = 'Transformer')
```

--------------------------------

### Initialize the UnitTest object

Source: https://github.com/jaidedai/easyocr/blob/master/unit_test/demo.ipynb

Creates an instance of the UnitTest class with specified paths and directories.

```python
unit_test = UnitTest(easyocr_module, 
                     test_data,
                     image_data_dir
                     )
```

--------------------------------

### Check CUDA and NVCC Version

Source: https://github.com/jaidedai/easyocr/blob/master/easyocr/DBNet/README.md

Verify if NVCC and CUDA toolkit are installed on your system. This is a prerequisite for compiling the CUDA version of DCN operators.

```bash
> nvcc --version
```

--------------------------------

### Initialize Reader with Multiple Languages

Source: https://context7.com/jaidedai/easyocr/llms.txt

Demonstrates initializing the EasyOCR Reader with different language groups. The library automatically selects the appropriate pretrained models based on the provided ISO 639 language codes.

```python
import easyocr

# Latin script: af az bs cs cy da de en es et fr ga hr hu id is it ku la lt lv
#               mi ms mt nl no oc pi pl pt ro rs_latin sk sl sq sv sw tl tr uz vi
reader = easyocr.Reader(['en', 'de', 'fr'])

# Arabic script: ar fa ug ur
reader = easyocr.Reader(['ar', 'en'])

# Cyrillic: ru rs_cyrillic be bg uk mn (+ minority languages)
reader = easyocr.Reader(['ru', 'en'])

# Devanagari: hi mr ne bh mai ang bho mah sck new gom sa bgc
reader = easyocr.Reader(['hi', 'en'])
```

--------------------------------

### Initialize Reader with Specific Recognition Model

Source: https://github.com/jaidedai/easyocr/blob/master/releasenotes.md

Specify a particular recognition model when initializing the EasyOCR Reader. For example, to use the 1st generation Latin model.

```python
reader = easyocr.Reader(['en','fr'], recog_network = 'latin_g1')
```

--------------------------------

### Initialize EasyOCR Reader with Bengali and English

Source: https://context7.com/jaidedai/easyocr/llms.txt

Instantiate the EasyOCR Reader with a list of desired languages. English is compatible with all languages.

```python
reader = easyocr.Reader(['bn', 'en'])
```

--------------------------------

### Handle incompatible language combinations in EasyOCR

Source: https://context7.com/jaidedai/easyocr/llms.txt

Scripts incompatible with each other cannot be mixed in one Reader. For example, Chinese and Arabic will raise a ValueError. This snippet demonstrates how to catch and print the error.

```python
try:
    reader = easyocr.Reader(['ch_sim', 'ar'])  # Raises ValueError
except ValueError as e:
    print(e)
```

--------------------------------

### Initialize EasyOCR Reader with CJK and other scripts

Source: https://context7.com/jaidedai/easyocr/llms.txt

Instantiate the EasyOCR Reader with various scripts like Simplified Chinese, Traditional Chinese, Japanese, Korean, Thai, Tamil, Telugu, and Kannada, along with English.

```python
reader = easyocr.Reader(['ch_sim', 'en'])   # Simplified Chinese
```

```python
reader = easyocr.Reader(['ch_tra', 'en'])   # Traditional Chinese
```

```python
reader = easyocr.Reader(['ja', 'en'])       # Japanese
```

```python
reader = easyocr.Reader(['ko', 'en'])       # Korean
```

```python
reader = easyocr.Reader(['th', 'en'])       # Thai
```

```python
reader = easyocr.Reader(['ta', 'en'])       # Tamil
```

```python
reader = easyocr.Reader(['te', 'en'])       # Telugu
```

```python
reader = easyocr.Reader(['kn', 'en'])       # Kannada
```

--------------------------------

### Basic OCR Usage

Source: https://github.com/jaidedai/easyocr/blob/master/README.md

Initialize the EasyOCR Reader with desired languages and perform text recognition on an image. The Reader initialization loads models into memory and should be done once.

```python
import easyocr
reader = easyocr.Reader(['ch_sim','en']) # this needs to run only once to load the model into memory
result = reader.readtext('chinese.jpg')
```

--------------------------------

### EasyOCR Command Line Usage

Source: https://github.com/jaidedai/easyocr/blob/master/README.md

Execute EasyOCR from the command line, specifying languages, input file, detail level, and GPU usage.

```shell
$ easyocr -l ch_sim en -f chinese.jpg --detail=1 --gpu=True
```

--------------------------------

### Initialize EasyOCR Reader

Source: https://context7.com/jaidedai/easyocr/llms.txt

Initialize the OCR engine by loading text detector and recognition models. Reuse the same `Reader` instance for multiple calls. Supports various configurations for language, GPU usage, model storage, and network types.

```python
import easyocr

# Basic: English only, GPU enabled (auto-detects CUDA or MPS)
reader = easyocr.Reader(['en'])
```

```python
# Multi-language: Simplified Chinese + English, CPU only
reader = easyocr.Reader(['ch_sim', 'en'], gpu=False)
```

```python
# Custom model storage path, no auto-download
reader = easyocr.Reader(
    lang_list=['en', 'de', 'fr'],
    gpu=True,
    model_storage_directory='/opt/ocr_models',
    download_enabled=True,
    detect_network='craft',      # 'craft' (default) or 'dbnet18'
    recog_network='standard',    # 'standard' or a custom model name
    quantize=True,               # dynamic quantization on CPU
    cudnn_benchmark=False,
    verbose=True
)
```

```python
# Custom recognition model (requires yourmodel.pth, .yaml, .py in user_network_directory)
reader = easyocr.Reader(
    lang_list=['en'],
    user_network_directory='~/.EasyOCR/user_network',
    recog_network='custom_example'
)
```

--------------------------------

### Integrate Custom Recognition Model

Source: https://context7.com/jaidedai/easyocr/llms.txt

Shows how to integrate a custom recognition model by specifying the directories for model weights and user network definitions. Ensure the 'recog_network' matches the model's filename stem.

```python
# File layout (example: ~/.EasyOCR/):
#   model/my_model.pth               ← trained weights
#   user_network/my_model.py         ← network class definition
#   user_network/my_model.yaml       ← configuration

# Example my_model.yaml:
# lang_list: ['en']
# character_list: "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"
# imgH: 64
# network_params:
#   input_channel: 1
#   output_channel: 256
#   hidden_size: 256

import easyocr

reader = easyocr.Reader(
    lang_list=['en'],
    model_storage_directory='~/.EasyOCR/model',
    user_network_directory='~/.EasyOCR/user_network',
    recog_network='my_model',   # matches the filename stem
    download_enabled=False      # weights are already local
)

result = reader.readtext('test_image.jpg', detail=1)
for bbox, text, conf in result:
    print(f"{text!r}  conf={conf:.3f}")
```