infer_mmlab_kie
Ikomia HUB

infer_mmlab_kie

GitHub

About

2.0.0
Apache-2.0

Inference for MMOCR from MMLAB KIE models

Task: OCR
infer
key
information
extraction
kie
mmlab
sdmgr

Run KIE (Key Information Extraction) algorithms from MMLAB framework. This algorithm will be applied after text detection and text recognition. You can use infer_mmlab_text_detection and infer_mmlab_text_recognition from Ikomia HUB for this task.

Models will come from MMLAB's model zoo if custom training is disabled. If not, you can choose to load your model trained with algorithm train_mmlab_kie from Ikomia HUB.

Example image

🚀 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

from ikomia.dataprocess.workflow import Workflow
from ikomia.utils.displayIO import display

# Init your workflow
wf = Workflow()

# Add algorithms...
# for text detection
det = wf.add_task(name="infer_mmlab_text_detection", auto_connect=True)
# for text recognition
rec = wf.add_task(name="infer_mmlab_text_recognition", auto_connect=True)
# for kie
kie = wf.add_task(name="infer_mmlab_kie", auto_connect=True)

# Run on your image
wf.run_on(url="https://github.com/open-mmlab/mmocr/blob/main/demo/demo_kie.jpeg?raw=true")

# Get results
original_image_output = kie.get_output(0)
text_detection_output = kie.get_output(1)

# Display results
display(original_image_output.get_image_with_graphics(text_detection_output))

☀️ 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

from ikomia.dataprocess.workflow import Workflow
from ikomia.utils.displayIO import display

# Init your workflow
wf = Workflow()

# Add algorithms...
# for text detection
det = wf.add_task(name="infer_mmlab_text_detection", auto_connect=True)
# for text recognition
rec = wf.add_task(name="infer_mmlab_text_recognition", auto_connect=True)
# for kie
kie = wf.add_task(name="infer_mmlab_kie", auto_connect=True)

# Set parameters
kie.set_parameters({
    'model_name': 'sdmgr', 
    'cfg': 'sdmgr_unet16_60e_wildreceipt'})

# Run on your image
wf.run_on(url="https://github.com/open-mmlab/mmocr/blob/main/demo/demo_kie.jpeg?raw=true")

# Get results
original_image_output = kie.get_output(0)
text_detection_output = kie.get_output(1)
  • model_name (str, default="satrn"): model name.
  • cfg (str, default="satrn_shallow-small_5e_st_mj"): name of the model configuration file.
  • config_file (str, optional): path to model config file (only if custom_training=True). The file is generated at the end of a custom training. Use algorithm train_mmlab_text_recognition from Ikomia HUB to train custom model.
  • model_weight_file (str, optional): path to model weights file (.pt) (only if custom_training=True). The file is generated at the end of a custom training.
  • dict_file (str, default="dicts/english_digits_symbols.txt"): characters dictionary. Set it when you use a custom train.
  • class_file (str, default="wildreceipt/class_list.txt"): Class list. Set it when you use a custom train.
  • merge_box (bool, default=True): Merge text boxes before running KIE algorithm.
  • max_x_dist (int, default=-1): Used if merge_box is True. Text boxes closer (on x-axis) than this value are merged. If max_x_dist est lower than 0, it will automatically calculate this value based on the mean height of all text boxes in the input. It will first perform a merging with a maximum distance equal to a quarter of mean height, joining boxes with '', then perform a second merging with maximum distance equal to mean height, joining boxes with ' '.
  • min_y_overlap_ratio (float, default=0.6): Used if merge_box is True. Text boxes can be merged if they overlap on y-axis more than this ratio.

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())

🔍 Explore algorithm outputs

Every algorithm produces specific outputs, yet they can be explored them the same way using the Ikomia API. For a more in-depth understanding of managing algorithm outputs, please refer to the documentation.

from ikomia.dataprocess.workflow import Workflow

# Init your workflow
wf = Workflow()

# Add algorithms...
# for text detection
det = wf.add_task(name="infer_mmlab_text_detection", auto_connect=True)
# for text recognition
rec = wf.add_task(name="infer_mmlab_text_recognition", auto_connect=True)
# for kie
kie = wf.add_task(name="infer_mmlab_kie", auto_connect=True)

# Run on your image
wf.run_on(url="https://github.com/open-mmlab/mmocr/blob/main/demo/demo_kie.jpeg?raw=true")

# Iterate over outputs
for output in kie.get_outputs():
    # Print information
    print(output)
    # Export it to JSON
    output.to_json()

MMLab text recognition algorithm generates 2 outputs:

  1. Forwarded original image (CImageIO)
  2. Text detection output (CTextIO)

Developer

  • Ikomia
    Ikomia

License

Apache License 2.0
Read license full text

A permissive license whose main conditions require preservation of copyright and license notices. Contributors provide an express grant of patent rights. Licensed works, modifications, and larger works may be distributed under different terms and without source code.

PermissionsConditionsLimitations

Commercial use

License and copyright notice

Trademark use

Modification

State changes

Liability

Distribution

Warranty

Patent use

Private use

This is not legal advice: this description is for informational purposes only and does not constitute the license itself. Provided by choosealicense.com.