Implementation:Open compass VLMEvalKit CCOCR Doc Parsing Evaluator

Field	Value
source	VLMEvalKit
domain	Vision, Evaluation, OCR, Document Parsing, Table Recognition

Overview

Evaluates document parsing quality using tree edit distance (APTED) for table structure recognition and LaTeX cleanup for document content in the CCOCR benchmark.

Description

This module implements `TableTree` and `CustomConfig` classes (adapted from IBM's work) for computing tree edit distance between predicted and ground-truth HTML table structures using the APTED algorithm. It includes LaTeX document preprocessing via regex pattern removal, NLTK-based tokenization, and TEDS (Tree Edit Distance based Similarity) scoring. The evaluator handles both table structure evaluation and general document parsing metrics.

Usage

Called internally by the corresponding dataset class during evaluation.

Code Reference

• Source: vlmeval/dataset/utils/ccocr_evaluator/doc_parsing_evaluator.py, Lines: L1-256
• Import: from vlmeval.dataset.utils.ccocr_evaluator.doc_parsing_evaluator import TableTree, CustomConfig

Key Functions:

class TableTree(Tree): ...
class CustomConfig(Config):
    def rename(self, node1, node2): ...

I/O Contract

Direction	Description
Inputs	Predicted and ground-truth HTML table strings or LaTeX document content
Outputs	TEDS similarity score (0-1) for table structures; text-level metrics for document content

Usage Examples

from vlmeval.dataset.utils.ccocr_evaluator.doc_parsing_evaluator import TableTree

tree = TableTree(tag="table")

Related Pages

• Principle:Open_compass_VLMEvalKit_Benchmark_Dataset_Construction