* pyclipper inhomogeneous expanded array solved
For some images, `np.array(offset.Execute(distance))` can result in inhomogeneous part of the detection box list, which cannot be casted into numpy array directly.
* corrected box reshape position
- box reshape was mistakenly done at line 145 which is now correctly done at line 92 of `db_postprocess.py`
- if box is empty then continue
* reverted mistakenly changed line 147
- reverted mistakenly changed `box.array(box)` to `np.array(box)`
* expanded array fix for `det_box_type=quad`
* polygons padding
For `--det_box_type = poly`, pad the detected polygon arrays if they have different shapes to ensure even shapes of polygon arrays
* fix codestyle
---------
Co-authored-by: Wang Xin <xinwang614@gmail.com>
* Fix the bug where Python scripts fail to execute PDF text recognition tasks, optimize the logic of judging PDF files, and add cases to the quickstart document for layout analysis.
* Add two examples of PDF layout analysis to the quickstart file of ppstructure.
* Add a return comment for the check_img function
* docs: Update FAQ.md, delete repeated question
* docs: 1.update the FAQ.md from the doc_ch, delete repeated question 2. update the FAQ_en.md from the doc_en, add questions and answers about "How to identify artistic fonts in signs or advertising images"
* docs: Update the FAQ.md from the doc_ch, delete repeated question
* docs: Update the FAQ.md from the doc_ch, delete repeated question
* Update quickstart_en.md
sync quickstart cn doc's better pdf demo
* Update quickstart.md
revert font location changes of the demo code
* Update quickstart_en.md
revert font location changes of the en demo code
fix issues:
1.getPixmap() function is not recognized,changing to get_pixmap
2.fix TypeError when paddle recognized an empty page
3.pre-stored pageCount to avoid issues
4.added GPU usage
This commit addresses the incorrect usage of the `try_import` function from `paddle.utils` in both `ppocr/utils/utility.py` and `ppstructure/pdf2word/pdf2word.py`.
* Handle conflict where a box is simultaneously recognized as multiple labels
* Split large height image recursively and process each with overlap to enhance performance
* Fix error when dt_box result is empty
* Add split operation on horizon side
* Slide on horizon may suffer line completeness, so that add more strict condition.
* Optimize recognition of overlap boxes.
jzhang533
张春乔
Wang Xin
Muhammad Asif
Mattheliu
Sanjay Rijal
Miaomiao Zhao
topduke
tackhwa
dyning
NOEXIST
Ichimaru Gin
S M
xu
wanghuancoder
Luo Peng
Eric Guo
Eric Guo
xiaoting
NeterOster
zxcd
Matej Kollár
Ran chongzhi
co63oc
sheiy
zhangyubo0722
firmament2008
Yesir
Weihang Wang
marswen