[Docs] How to use TextDetInferencer

[zfxw]

Branch

master branch https://github.com/open-mmlab/mmocr

📚 The doc issue

I trained a Textdet model with my own data, I can see a good detection result with following command
python tools/test.py ~/mmocr/work_dirs/dbnet_resnet18_fpnc_1200e_blfsh/dbnet_resnet18_fpnc_1200e_blfsh.py ~/mmocr/work_dirs/dbnet_resnet18_fpnc_1200e_blfsh/epoch_400.pth --show-dir imgs/.

But when I tried to use TextDetInferencer with following code

from mmocr.apis import TextDetInferencer

# Load models into memory
infer = TextDetInferencer(model='<Home>/mmocr/work_dirs/dbnet_resnet18_fpnc_1200e_blfsh/dbnet_resnet18_fpnc_1200e_blfsh.py', weights='<Home>/mmocr/work_dirs/dbnet_resnet18_fpnc_1200e_blfsh/epoch_400.pth', scope='mmocr')

# Perform inference
result = infer('<Home>/BLFSH_training_data/OpenMM_textbox_training_data/test/202402020_124623_cr02.png')
print(result)

All the bounding boxes are out of image bounds, and here is the error:

10/10 21:06:44 - mmengine - WARNING - Failed to search registry with scope "mmocr" in the "function" registry tree. As a workaround, the current "function" registry in "mmengine" is used to build instance. This may cause unexpected failure when running the built modules. Please check whether "mmocr" is a correct scope, or whether the registry is initialized.
/opt/conda/miniconda3/lib/python3.8/site-packages/mmengine/visualization/visualizer.py:196: UserWarning: Failed to add <class 'mmengine.visualization.vis_backend.LocalVisBackend'>, please provide the `save_dir` argument.
  warnings.warn(f'Failed to add {vis_backend.__class__}, '
/opt/conda/miniconda3/lib/python3.8/site-packages/mmengine/visualization/visualizer.py:831: UserWarning: Warning: The polygon is out of 
bounds, the drawn polygon may not be in the image

Why the polygon could be out of the bounds? Am I using TextDetInferencer properly?

I really appreciate if you can help. Thank you

Suggest a potential alternative/fix

No response

[yychayitu]

Have you solved this problem? I meet the same problem in test.py

[zfxw]

I do solve the problem, but because it was long time ago, I can't remember what exactly happened, I remember it was because something wrong when I generate the training data.

[yychayitu]

Thank you for your suggestions. I will try.