use default ser ckpt in vi_layoutxlm_ser

[hongziqi]

Thank you for your contribution to the MindOCR repo.
Before submitting this PR, please make sure:

• You have read the Contributing Guidelines on pull requests
• Your code builds clean without any errors or warnings
• You are using approved terminology
• You have added unit tests

Motivation

1. update register_model: vi_layoutxlm_ser to use pretrain ckpt
2. fix bug by inferring ser in layoutxlm_ser:
   [ERROR] ANALYZER(2921322,ffffa6185010,python):2024-12-30-08:59:24.113.844 [mindspore/ccsrc/pipeline/jit/ps/static_analysis/evaluator.cc:713] Run] Primitive: <StandardPrimEvaluator_list_getitem> infer failed, failed info: list_getitem evaluator index should be in range[-4, 4), but got 4.

Before Update（would not load ckpt in mindocr/mindocr/models/kie_layoutxlm.py）

1. infer with vi_layoutxlm:

python tools/infer/text/predict_ser.py --rec_algorithm CRNN_CH --det_algorithm DB_PPOCRv3 --image_dir configs/kie/vi_layoutxlm/example.jpg

2. infer with layoutxlm, the following bugs have occurred:

python tools/infer/text/predict_ser.py --rec_algorithm CRNN_CH --det_algorithm DB_PPOCRv3 --image_dir configs/kie/vi_layoutxlm/example.jpg --ser_algorithm LAYOUTXLM

After Update（use the best ckpt link）

1. infer with vi_layoutxlm:

python tools/infer/text/predict_ser.py --rec_algorithm CRNN_CH --det_algorithm DB_PPOCRv3 --image_dir configs/kie/vi_layoutxlm/example.jpg

2. infer with layoutxlm:

python tools/infer/text/predict_ser.py --rec_algorithm CRNN_CH --det_algorithm DB_PPOCRv3 --image_dir configs/kie/vi_layoutxlm/example.jpg --ser_algorithm LAYOUTXLM

3. eval with vi_layoutxlm:

python tools/eval.py --config configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yaml

4. eval with layoutxlm:

python tools/eval.py --config configs/kie/layoutxlm/ser_layoutxlm_xfund_zh.yaml

5. train with vi_layoutxlm:

python tools/train.py --config configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yaml

6. train with layoutxlm:

python tools/train.py --config configs/kie/layoutxlm/ser_layoutxlm_xfund_zh.yaml

Test Plan

test for

# download mindocr
git clone -b feat-update-kie-ckpt-link https://github.com/hongziqi/mindocr.git
# build mindocr wheel package and install mindocr
python setup.py bdist_wheel
pip install dist/mindocr-0.4.0-py3-none-any.whl

# infer with VI_LAYOUTXLM
python tools/infer/text/predict_ser.py --rec_algorithm CRNN_CH --det_algorithm DB_PPOCRv3 --image_dir configs/kie/vi_layoutxlm/example.jpg

# infer with LAYOUTXLM
python tools/infer/text/predict_ser.py --rec_algorithm CRNN_CH --det_algorithm DB_PPOCRv3 --image_dir configs/kie/vi_layoutxlm/example.jpg --ser_algorithm LAYOUTXLM

# download dataset
wget https://download.mindspore.cn/toolkits/mindocr/vi-layoutxlm/XFUND.tar && tar -xf XFUND.tar
# train or eval (need to update the path in yaml, please refer to the file: configs/kie/vi_layoutxlm/README_CN.md)
python tools/train.py(eval.py) --config configs/kie/vi_layoutxlm/ser_vi_layoutxlm_xfund_zh.yaml
python tools/train.py(eval.py) --config configs/kie/layoutxlm/ser_layoutxlm_xfund_zh.yaml

[CaitinZhao]

训练脚本是不是也要适配

[hongziqi]

不影响训练脚本，训练配置的是backbone的pretrained参数，这里并没有改动backbone网络的参数。