train_mmlab_kie

Training process for MMOCR from MMLAB in Key Information Extraction (KIE).

🚀 Use with Ikomia API

1. Install Ikomia API

We strongly recommend using a virtual environment. If you're not sure where to start, we offer a tutorial here.

pip install ikomia

2. Create your workflow

To try this code snippet, you can download and extract from wildreceipt. Then make sure you fill the parameter dataset_folder correctly.

from ikomia.dataprocess.workflow import Workflow

# Init your workflow
wf = Workflow()

# Add dataset
dataset = wf.add_task(name="dataset_wildreceipt", auto_connect=False)

# Set dataset parameters
dataset.set_parameters({'dataset_folder': 'dataset/folder'})

# Add algorithm
algo = wf.add_task(name="train_mmlab_kie", auto_connect=True)

# Set parameters
algo.set_parameters({'batch_size': 2})

# Run training
wf.run()

📝 Set algorithm parameters

• model_name (str, default="sdmgr"): Model name. Set model_name and cfg to choose which model to train. See code snippet below to know what are the possibilities.
• cfg (str, default="sdmgr_novisual_60e_wildreceipt.py"): Config.
• pretrain (bool, default=True): Use pretrained model from MMLAB. To train a model from scratch, set this to False.
• epochs (int, default=10): number of complete passes through the training dataset.
• batch_size (int, default=4): number of samples processed before the model is updated.
• dataset_split_ratio (int, default=90): in percentage, divide the dataset into train and evaluation sets ]0, 100[.
• output_folder (str): path to where the model will be saved. Default folder is "runs/" in the algorithm directory.
• eval_period (int, default=1): interval between evaluations.
• dataset_folder (str): path to where the dataset compatible with mmlab is stored. Default folder is "/dataset" in the algorithm directory.
• expert_mode (bool, default=False): set to True only if you know how mmlab works. Then you can set all the parameters in the mmlab config system and it will override every other parameters above.
• config_file (str, default=""): path to the .py config file. Only for custom models.
• model_weight_file (str, default=""): path to the .pth weight file. Only for custom models.

Note: parameter key and value should be in string format when added to the dictionary.

MMLab framework offers multiple models. To ease the choice of couple (model_name/cfg), you can call the function get_model_zoo() to get a list of possible values.

from ikomia.dataprocess.workflow import Workflow

# Init your workflow
wf = Workflow()

# Add kie algorithm
kie = wf.add_task(name="infer_mmlab_kie", auto_connect=True)

# Get list of possible models (model_name, model_config)
print(kie.get_model_zoo())

☀️ Use with Ikomia Studio

Ikomia Studio offers a friendly UI with the same features as the API.

• If you haven't started using Ikomia Studio yet, download and install it from this page.

• For additional guidance on getting started with Ikomia Studio, check out this blog post.