Edit on GitHub
  • John Snow Labs
  • Mar 15, 2024

Table Cell Detection v2

  • en
  • licensed

Description

Object detection model trained to detect table cells one of the foremost architectures in the state-of-the-art, meticulously selected through benchmark evaluations and comparative analyses. Trained on an extensive and diverse dataset, this model has been finely tuned for precise table cell detection within documents. Its efficacy has been verified through rigorous testing, demonstrating exceptional performance in table cell detection across a spectrum of document formats.

Predicted Entities

cells.

Live Demo Open in Colab Copy S3 URI

How to use

binary_to_image = BinaryToImage()

img_to_hocr = ImageToHocr() \
    .setInputCol("image") \
    .setOutputCol("hocr") \
    .setIgnoreResolution(False) \
    .setOcrParams(["preserve_interword_spaces=0"])

cell_detector = ImageDocumentRegionDetector() \
    .pretrained("region_cell_detection_v2", "en", "clinical/ocr") \
    .setInputCol("image") \
    .setOutputCol("cells") \
    .setScoreThreshold(0.8)

draw_regions = ImageDrawRegions() \
    .setInputCol("image") \
    .setInputRegionsCol("cells") \
    .setOutputCol("image_with_regions") \
    .setRectColor(Color.red)

hocr_to_table = HocrToTextTable() \
    .setInputCol("hocr") \
    .setRegionCol("table_regions") \
    .setOutputCol("tables") \
    .setUseCellsCol("cells")

pipeline = PipelineModel(stages=[
    binary_to_image,
    img_to_hocr,
    cell_detector,
    draw_regions,
    hocr_to_table
])

imagePath = 'resources/ocr/table.jpg'
image_df = spark.read.format("binaryFile").load(imagePath).sort("path")

result = pipeline.transform(image_df)

val binary_to_image = BinaryToImage()

val img_to_hocr = ImageToHocr()
    .setInputCol("image")
    .setOutputCol("hocr")
    .setIgnoreResolution(False)
    .setOcrParams(Array("preserve_interword_spaces=0"))

val cell_detector = ImageDocumentRegionDetector()
    .pretrained("region_cell_detection_v2", "en", "clinical/ocr")
    .setInputCol("image")
    .setOutputCol("cells")
    .setScoreThreshold(0.8)

val draw_regions = ImageDrawRegions()
    .setInputCol("image")
    .setInputRegionsCol("cells")
    .setOutputCol("image_with_regions")
    .setRectColor(Color.red)

val hocr_to_table = HocrToTextTable()
    .setInputCol("hocr")
    .setRegionCol("table_regions")
    .setOutputCol("tables")
    .setUseCellsCol("cells")

val pipeline = new PipelineModel().setStages(Array(
    binary_to_image,
    img_to_hocr,
    cell_detector,
    draw_regions,
    hocr_to_table))

val imagePath = "resources/ocr/table.jpg"
val image_df = spark.read.format("binaryFile").load(imagePath).sort("path")

val result = pipeline.transform(image_df)

Example

Input:

Screenshot

Cell Detection:

Screenshot

Table Structure Recognition:

Screenshot

Model Information

Model Name: region_cell_detection_v2
Type: ocr
Compatibility: Visual NLP 5.2.0+
License: Licensed
Edition: Official
Language: en