train_mmlab_text_detection

Train text detection models from MMLAB.

🚀 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 text recognition dataset
dataset = wf.add_task(name="dataset_wildreceipt", auto_connect=False)

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

# Add train algorithm
train = wf.add_task(name="train_mmlab_text_detection", auto_connect=True)

# Set train algorithm parameters
train.set_parameters({'model_name': 'dbnetpp', 
                      'cfg': 'dbnetpp_resnet50-dcnv2_fpnc_1200e_icdar2015',
                      'epochs': '10',
                      'batch_size': '2'})

# Launch training
wf.run()

☀️ 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.

📝 Set algorithm parameters

• model_name (str, default="dbnet"): name of pretrained model.
• cfg (str, default="dbnet_resnet18_fpnc_1200e_icdar2015.py"): filename of pretrained model's config.

model_name and cfg work by pair. You can print the available possibilities with this code snippet:

from ikomia.dataprocess.workflow import Workflow

# Init your workflow
wf = Workflow()

# Add algorithm
algo = wf.add_task(name="train_mmlab_text_detection")

# Get model zoo and print it
model_zoo = algo.get_model_zoo()
print(model_zoo)

# Set parameters with the first model of the list
algo.set_parameters(model_zoo[0])

• 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.

from ikomia.dataprocess.workflow import Workflow

# Init your workflow
wf = Workflow()

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

algo.set_parameters({
    "model_name": "dbnetpp",
    "cfg": "dbnetpp_resnet50_fpnc_1200e_icdar2015.py",
    "epochs": "20",
    "batch_size": "2",
    "eval_period": "2",
    "dataset_split_ratio": "90",
    "output_folder": "/out",
    "dataset_folder": "/dataset",
    "export_mode": "False",
    "config_file": "",
    "model_weight_file": ""
})

# Continue your workflow