Unable to Complete vis_match on Large Datasets (>10,000 LaTeX Formulas)

[BFlameSwift]

Hello, I am encountering an issue when using the Character Detection Matching (CDM) metric as described in your paper "CDM: A Reliable Metric for Fair and Accurate Formula Recognition Evaluation".

The evaluation runs smoothly for small datasets (e.g., 1,000 LaTeX formulas). However, when processing larger datasets (>10,000 formulas), the evaluation fails to complete thevis match step, preventing proper comparison of results. Below is the error output:

2024-11-25 11:24:37 extract bbox done, time cost: 645.438 s
100%|██████████████████████████████████████████████████████████████████| 24600/24600 [00:00<00:00, 189073.52it/s]
/home/user1/miniconda3/envs/cdm_1120/lib/python3.9/site-packages/numpy/core/fromnumeric.py:3504: RuntimeWarning: Mean of empty slice.
  return _methods._mean(a, axis=axis, dtype=dtype,
/home/user1/miniconda3/envs/cdm_1120/lib/python3.9/site-packages/numpy/core/_methods.py:129: RuntimeWarning: invalid value encountered in scalar divide
  ret = ret.dtype.type(ret / rcount)
2024-11-25 11:24:38 calculate metrics done, time cost: 0.151 s
=> process done, mean f1 score: nan

Steps to Reproduce:

1. Use the provided CDM implementation.
2. Prepare a dataset of over 10,000 LaTeX formulas.
3. Run the evaluation pipeline, ensuring the vis_match step is included.

Expected Behavior:

The vis_match step should complete successfully, and the evaluation metrics (e.g., mean F1 score) should be computed without errors.

Actual Behavior:

The evaluation halts during the vis_match step, resulting in an nan (not a number) mean F1 score.

Environment:

Operating System: Ubuntu 20.04
Python Version: 3.9

I suspect the issue might be related to memory usage or internal handling of bounding boxes for large datasets. Any guidance or suggestions on addressing this issue would be greatly appreciated.