PGNet model downloaded does not perform as good as reported.

[lidan1]

问题描述 / Problem Description

I tested the PGNet downloaded from the link provided in the doc on TotalText test set, the numbers I got was lower than reported in the table.
https://github.com/PaddlePaddle/PaddleOCR/blob/main/doc/doc_en/algorithm_e2e_pgnet_en.md#test-environment-nvidia-tesla-v100-sxm2-16gb (url)

运行环境 / Runtime Environment

• OS: Linux
• Paddle: 2.4.0-rc0
• PaddleOCR: 2.4

复现代码 / Reproduction Code

`python3 tools/eval.py -c configs/e2e/e2e_r50_vd_pg.yml -o Global.checkpoints="{path/to/weights}/best_accuracy"

完整报错 / Complete Error Message

可能解决方案 / Possible solutions

附件 / Appendix

[wxsIcey]

hello，我使用相同数据集对PGNet也进行了评估，发现和你结果一致，请问这个问题得到解决了吗？